This page has been archived by the Computer Arts Society.

Case study: Harold Cohen and AARON

<<Previous Up Next>>

Whereas Hébert has moved towards a physical realisation for his art, Harold Cohen has moved away from it, to a degree. Cohen was a noted abstract painter during the 1960s, but following a retrospective at the Whitechapel in 1967 he became disenchanted with the British art scene and went to America to teach at UC San Diego. There, he was introduced to computers and gradually developed the program AARON, which has been the focus of his artistic activity for over twenty-five years. AARON developed from a fascination with the process of line-making and how enclosed forms, or shapes, were drawn on paper; initially Cohen did not approach his programming as an artistic activity, but rather a disciplined research into the grammar of conceptual space. He initially saw AARON as a program that emulated what humans did; then as an autonomous entity. As the program’s complexity increased, Cohen added more forms to its repertoire, eventually ending up with human figures and giving the program an increasingly sophisticated understanding of their positioning in space. Part of its fascination lay in the way that AARON’s images looked subjectively like sketches:

The earliest versions of AARON could do very little more than to distinguish between figure and ground, closed forms and open forms, and to perform various simple manipulations on those structures. […] AARON’s drawings had a distinctly freehand, ad-hoc look quite at odds with popular assumptions at that time about machines […]to judge from the responses of art-museum audiences, AARON’s marks quite evidently functioned as images[1]

The process of AARON’s picture-making is outlined by Ed Burton. The program receives its knowledge of the world indirectly, in that Cohen encodes the structures it uses to develop images: for instance, the human body as described in AARON is composed of parts in relative sizes, with a range of movement and posture and information on how they fit together. At a higher level, AARON can compose the objects in a scene such that they are positioned relative to each other. Burton emphasizes the tree-like structure of this process, and that the knowledge of structure is combined with AARON’s “procedural knowledge” of drawing, to arrive at a form.[2]

[Plate XXVII: AARON output]

The relation of rule to result is not as straightforward here as in, say, Helaman Ferguson’s sculptures which are solutions to specific equations; nor as in Hébert ’s algorist pieces. Cohen’s rules do not prescribe the exact form but rather its overall outcome: a series of branching “ifs” where results and outcome spawn yet more questions. His approach is heuristic and systems-based: not really mathematical. In this way, it is quite conceivable that no two AARON pictures could ever be alike, yet they can all share a similar visual style. To what extent is this style the outcome of Cohen’s direct programming, or is it somehow inherent in the picture-descriptions in ways its creator may not have anticipated. This reflects Cohen’s thoughts in a 1973 essay from the very beginning of his program:

Human art-making […] is characterized by a fluently changing pattern of decisions based on the artist’s awareness of the work in progress[3]

AARON could be described as an independently creative extension of Cohen’s artistic method, rendered into software and executed by the computer. Insofar as AARON can create pictures, it does so in a style that has emerged over the course of Cohen’s programming and development of AARON’s technique. In its development of Artificial Intelligence theories, AARON also shows how a relatively simple set of rules can lead to complex and intelligible results; an artistic emergence remarked upon by McCorduck:

In AARON, a central idea of artificial intelligence is exemplified: the program is able to generate the illusion of a complete and coherent set of images out of a comparatively simple and sparse lower representation. [4]

Of particular interest is Cohen’s notion that the AARON pictures represent a “collaboration” between himself and the program. The forms incorporated into its core have evolved from his understanding of the art-making process, and their structuring into higher forms and scenes also represents his input. Cohen’s agency is at the instructional end, but rather than issue specifications (or orders, like Moholy-Nagy) he gives outlines, patterns to be incorporated as the program sees fit. (Must ask if it operates on an “evolutionary” algorithm of fitness) This is what McCorduck refers to when she sets forth a few of the rules. Yet the program’s size implies these rules are exhaustively described. AARON’s room for manoeuvre – “creativity” if you will – comes from the application of these rules in visual form.

Cohen posed me the example of facial construction: the nose can vary within a plausible range, as with the eyes, etc; all considerations taken together still allows an infinite range of possible faces. If this plausibility factor could be deduced by the machine, this would move it somewhat towards autonomy. [5] When the program is about to be executed, HC changes the variables within the code and the program begins to draw, allowing space for overlapping parts of the picture in its initial image block.  However, the problem is that due to the size of the program, modifications to one part can seriously affect the others.

This was graphically demonstrated when HC modified one command to fill in the irises in the eyes of figures drawn by AARON: the net result was that the program drew up to this point, then crashed. It would seem the modification necessitated changes in other places to make it work. Changing these other parameters required Cohen to work on parts of the program he had not seen for up to a year: with so many lines of code, he had to return to his own knowledge of its workings and think back to his previous modifications and changes. In this way, the program is an integral part of him, and his art, and it would be hard to separate the two: AARON is an external manifestation of how he conceives of and executes his art.

Although the resultant picture is in no way predictable, it does contain certain features that mark it out as being an “AARON” work, many of which have emerged in the course of Cohen’s programming and modification. Indeed, Cohen goes so far as to say the program will continue to turn out “originals” long after his death, enabling him to have a posthumous show of new works! However, Larry Cuba rather derisively termed AARON a “Harold Cohen simulator”, pointing out that in his opinion Cohen was only modelling his own internalised drawing process; any thought of its universal applicability shout be regarded sceptically.[6] In fairness, Cohen is quick to acknowledge this aspect of his work.[7] So, is AARON simply a high-level mechanism with a high degree of seeming artistry, or does it span the gap into a truly creative computer artist that can produce visually meaningful (as opposed to intentionally random) results?

The output of AARON’s pictures is interesting in itself: it began with pictures drawn on paper with plotters and a turtle, and has evolved to include oil paints along with a knowledge of colour. Cohen stopped using the painting machines because of their engaging nature and the way that people were requesting pictures of them, confusing the external manifestation with the AARON software. The turtle he once used for drawing pictures was also very involving. The machines’ drawing process had the feel of a performance, and people were delighted when the painting machine washed up its own cups. Cuba speculates that the appeal of Cohen was seeing the plotter creating the drawing in the gallery.

References to the “robot” artist annoyed Cohen because its physical existence is beside the point. It was only intended as an exhibition device, not as the focus of the art. AARON as a program embodies a process, not a physical machine.[8] In its latest incarnation, however, a version of AARON as a screensaver for the PC is being distributed by Ray Kurzweil as a trial download from his website.

Cohen contends that by offering AARON across the Internet, he is not only extending his previous premise of enabling everyone to own the means to make art, but he has also cloned the artist, in the sense that he has reproduced an art-making process. He now claims that instead of mass-producing ART, he has cloned the artist. Every instance of every AARON program downloaded from Kurzweil’s website[9] will make different pictures forever. True, all the pictures are evidently the results of the same process, having a similar feel, but the fact they have any observable style is itself a consequence of Cohen’s approach. Cohen thinks the notion of “authorism” has outdated implications which have to be updated. AARON’s release as a software package has wiped out any value of the program’s uniqueness. Now anyone can have a copy, perhaps the value only remains in plottings and prints from HC’s own machine: those produced by other people will tend not to be included in the same subjective category.

On the other hand, Cohen’s collaborations with AARON also take the form of large-scale pictures, generated by the program and projected onto canvas which Cohen himself works up and paints. This is a more literal, physical interpretation of “collaboration”, going a stage further than the painting machine that once rendered AARON’s art. Cohen sees the canvases as permanent traces, or at least more long-term traces, of AARON’s activity than the drawings he used to make with the plotter or the ephemeral screen-based results of the AARON screensaver. He maintains, however, that the central importance of both activities – the canvas and the screen image – is that they are unique works of art: no two AARON works are exactly the same, and may this be regarded as originals.

By giving anyone with a PC the ability to make originals, Cohen claims he has turned Walter Benjamin’s idea of the mass-produced image on its head. Instead of copying the final image, with attendant notions of destroying the “aura” of the unique object, he has instead cloned the artist, or at least the process by which unique images are generated. Jean-Pierre Hébert has reservations about this last point: he sees AARON not so much as an artistic clone as a “mummy”, in the sense of an undead creature that has the appearance of life but no motivating spark.[10] It is certainly true that AARON operates as a bounded system, with only as much knowledge as Cohen is able to encode. Yet, As McCorduck says

[…] The program AARON, [Cohen] believes, stands in relation to its individual drawings the way a platonic ideal stands in relation to its earthly instantiations. […] Cohen has found a way to work his will upon and through the paradigm rather than upon a single instantiation simply means that his level of involvement is much higher, conceptually speaking, than has ever before been possible for the visual artist. [Since the program is responsible for the performances] it is as if a score could play itself [11]

If one considers Robin Baker’s criteria for a program to be to be recognised as creative, it seems that AARON satisfies the first point:

1) The conceptual space of the programmer is extended or
broken by a creative program. [In other words, it creates
something beyond the boundaries of what was originally

programmed into it.]

2) It should have judgment and be able to recognise its own

Undoubtedly, AARON has increased Cohen’s frame of reference, the extent of his art and his experiences with image making. However, his insistence on the collaborative aspects suggests that AARON will not fulfil the second point – even in all its multiple instances, the program will not become conscious of its artistic role, or even the relative merits of a “good” picture (one that its owner decides to keep) with a “bad” picture, which is cancelled. Nor could it ever extrapolate its own aesthetic rules.

[Plate XXVIII: AARON, Man]

Judged as a work of art (especially as a work of computer art, though Cohen would dispute that term), AARON presents the interesting spectacle of not merely rerunning the image set out in a program, but improvising upon a set of rules in different ways every time.

A further consideration with AARON is the legal status of the indirect work of art. It is interesting to note that the current Copyrights, Designs and Patents Act 1988 has provisos for As to the production of original pieces of art using computers, certain details in the 1988 Act relating to the artistic category of “photographs” may provide some clues to a satisfactory legal definition, with mechanical considerations in addition to the aesthetics:

Authors’ rights systems tend to give copyright only to “photographic works”, that is the results of careful and distinctive arrangement (scene-setting, lighting, angle, etc.), involving an element of aesthetic judgment which is personal to the photographer (and/or some “director”, rather than the mere cameraman). […] this will exclude not only casual snapshots […] but also press photography.[13]

The implication is that aesthetic consideration of artistic photography arises from the conjunction of several factors and is not always simply the result of the person holding the camera. This may carry over into computer “art”, where the computer’s construction of the image according to the artist’s direction has raised questions about the artist’s level of involvement. The 1988 Act considers the general category of “Computer-generated works”, as discussed by Cornish:

Computer-generated works

[…] the author of a literary, dramatic, musical or artistic work is its creator in a real sense. He or she (but not it) is the person who, by exercising labour, skill and judgment, gives expression to ideas of the appropriate kind. [Yet] the 1988 Act acknowledges that works of all these types may be computer-generated; and it provides that, where the circumstances are such that there is no human author of such a work, the author shall be taken to be the person by whom the arrangements necessary for creation of the work are undertaken.[14] [Emphasis mine]

This is deliberately vague, even more so than the preceding sections on artistic works. Does it recognise the artist per se, or the programmer, or even some patron who ordered the artwork? Subjectively, it seems sensible to recognise that the computer has no independent creative power and is only undertaking to execute an idea presented to it, or worked on using it, by the artist. However, in systems like those created by Harold Cohen, William Latham and others, where the computer executes images according to complex rules that make it seem to be acting independently, determining the “author” is almost impossible by traditional standards of the term. Harold Cohen’s AARON system gains its apparent artistic independence through Cohen’s skill at describing processes and habits of thought as rules that direct the system. He considers how this makes it different from straightforward computer graphics software:

AARON is an autonomous intelligent entity: not very autonomous, or very intelligent, or very knowledgeable, but very different, fundamentally different, from programs designed to be “just” tools. Electronic paint boxes, for example. And its use is equally different from the way computer artists use electronic paint boxes: I don’t work with the program, that is to say, I work on the program. The goal is the program’s autonomy, not the making of a better – orthodox – tool.[15]

Cohen, who once insisted on being the artist in all instances of AARON’s work, is now happy to share credit with his system.[16] Thus the legal ramifications of setting up an artistic system might be compared with the concept of a “work” in the musical sense:

The legally recognized work of art must have substance and must exist in the real world. But what type of physical and temporal existence does it have to have? What does the law do when confronted, for example, with music, which cannot be said to exist in the notes alone, nor in the performance, nor in the perception of the performance? [17]

It is at this point that one must consider the disjuncture between the digital information that underpins the computer image and the visual qualities of the image itself. Remembering Binkley’s thoughts, one could say that because the computer imposes no visual form in and of itself, it is not so much a medium as an instrument, albeit one that can “play” of its own accord. So the artistry lies in guiding the instrument’s movement. In this sense, the computer artist can achieve a similar position to a composer, who relies on others to perform his work. The legal arguments surrounding the computer artist should recognise this fact:

There must be overt behaviour manifesting substantial skill and/or labour which results in some form of detectable notation […] For example, in the case of dance the copyright law will recognize choreographic notation; in the case of architecture, architectural plans. Even for the visual arts such as painting, there is no reason to deny legal protection […] for the plans alone.[18]

In this way, computer art could be seen in a similar light to other artforms that cross the boundary between physical and symbolic, like dance and music. The instructions making up the computer image can only be executed by the computer, unless realised in material form in which case they have departed the computer’s realm and now exist independently of it, in the physical world. These considerations will be examined later, when the existence of computer artforms is discussed.

[1] Harold Cohen, Cohen, Harold “The Further Exploits of AARON, Painter” Stanford Electronic Humanities Review, volume 4, issue 2: “Constructions of the mind”. July 22, 1995.

[2] Ed Burton, “Representing Representation: Artificial Intelligence and Drawing” in Mealing, Stuart Computers and Art, (Exeter, 1997), p59

[3] Pamela McCorduck, AARON’s Code, p45

[4] McCorduck, ibid, p28

[5] Interview with Harold Cohen, August 2001

[6] Interview with Larry Cuba, August 2001

[7] Interview with Harold Cohen, 2001

[8] Interview with Harold Cohen, August 2001

[9] See Ray Kurzweil’s website:

[10] Correspondence with JP Hébert, December 2002

[11] Pamela McCorduck “Artificial Intelligence: An Aperçu”, in Neil Sieling, ed. The Technological Imagination: Machines in the Garden of Art (Minneapolis 1989) p22

[12] Robin Baker, Director of Ravensbourne, speaking at the conference “Beyond Art”, organised by Oxford HCU, 1999

[13] W.R. Cornish, ibid, p390.

[14] W.R. Cornish, ibid, p392.

[15] Harold Cohen “Off the shelf”, The Visual Computer (1986) 2 : 191 – 194c Springer – Verlag 1986

[16] Cohen, interview,

[17] Peter H. Karlen, “Art in the Law”, Leonardo, Vol.14, p.51

[18] Peter H. Karlen, “Art in the Law”, Leonardo, Vol.14, p.51