HAROLD COHEN: COMPUTER COLLABORATIONS

Silent Canyon 2, 2008. Pigment ink on panel, 117 x 191 cm

London’s Gazelli Art House is currently staging a retrospective of Harold Cohen (1928-2016). The was the first leading artist to take the possibilities of computer-made art seriously, Cohen combined that interest with a distinguished and related academic career – he was, for example, Director of the Center for Research in Computing and the Arts at UC San Diego from 1992-98.

After studying in London and Nottingham in the 1950’s, he developed a considerable reputation as an abstract painter, and represented Great Britain in the 1966 Venice Biennale. By then, seeking to move beyond subjective decision-making, he had begun to explore the possibilities for distancing himself from his own productions by devising a set of rules up front which he would then follow. So it was a natural fit when, shortly after he moved in 1968 to teach in San Diego, someone offered to introduce him to computer programming. Not that it was easy: as Cohen recalled fifty years later[i],  computing then ‘meant IBM cards chugging slowly through computers literally millions of times less powerful than today's laptops, and it could cost you hours, sometimes days, to determine that there was a semi-colon missing on one of your cards.’ Yet that did give Cohen a language in which rules could be expressed clearly, backed up by a machine that could actually execute the rules, and he went on to develop his AARON computer when invited to do research at Stanford University’s Artificial Intelligence Lab in 1973-75. The factors in play were the computer, an arm with implements to carry out the actions, and a table on which the ground to be painted was positioned. Typically, Cohen set things running overnight, so that he woke up to see what AARON – his ‘other self’ – had made while he slept.

Drawing Machine (Arm), 1980.

The image content was generated from code. Like Silent Canyon 2, much of AARON’s early material was, as Cohen put it, ‘essentially, though not explicitly, botanical; I mean that my goal was not to generate identifiable and ‘correct’ images of particular plants, but rather to generate material that could stand-in for those plants in evoking a natural environment. From a programming point of view, the code emulated the structure of leaves and trees, not their appearances, and each element would be grown from scratch, so to speak, not trotted out as variations of stored prototypes.’

Installation view at Gazelli Art House with drawings from 1983 and printing table.

Those early computers were, however, limited to black and white drawings – if Cohen wanted colour, he had to paint it in himself. Untitled Amsterdam Suite II – with its underwater suggestions – is an early AARON drawing coloured by Cohen, as is the monumentally-sized Untitled, 1981. He found that a little absurd: ‘if the program was so damned smart, why couldn't it colour its own drawings? Was it working for me or was I working for it?’

Untitled Amsterdam Suite 11, 1977. Coloured pencil over ink on paper 56.4 × 72.7 cm.

 But even when colour printing became possible, Cohen found himself dissatisfied with the computer’s ability to select the colours to be used, which he had programmed based on the human way of moving from one colour to another with modifications in response to newly-emergent relationships. It was a rule-based approach: if this then happens, respond that way. 1995’s Suite for Discovery Channel Online illustrates the results, which Cohen saw as the computer mimicking human processes rather than bringing a distinctive approach into play.

Installation view at Gazelli Art House with First Athletes, Athlete Series, 1986 (Acrylic on canvas 246.5 × 305 cm) and Untitled, 1981 (Acrylic on canvas 273 × 553 cm)

Only in 2005 did Cohen hit on an alternative. In any electronic display system, everything comes down to red, green and blue, and Cohen had been thinking in terms of how to programme their various possible mixtures. Then it occurred to him ‘to think more directly in terms of what their manipulation was supposed to accomplish; that is, if I could manipulate the hue, brightness and saturation directly.[ii]’ So he bypassed entirely the specification of individual colours that had characterized the earlier system, and allowed those colours to emerge from the relationships the program generated in terms of hue, lightness and saturation.

Jerry, 1992. Oil on canvas, 61 × 86.4 cm.

As he explained: ‘Hues were determined by subject matter. It didn’t much matter what hue the program chose for leaves or flowers, but the distances between the hues for different elements – flowers, tendrils, branches and so on – did matter and to fix those distances for AARON’s entire output, I settled on an additive series shifted to a random starting point. Every image would then have the same spacing of hues but no two of them would have the same hues or, consequently, the same colours. For the lightnesses and saturations to be associated with each of these hues, the program simply generated two lists of random values. These were then the only values that could be used for the entire image. As each plant element came up for colouring – leaf, flower, whatever - its hue would be whatever had been prescribed for it and its lightness and saturation would be drawn randomly from the respective list of values.’

Cohen found that limiting the lightness and saturation to a small set of fixed values provided coherence to the colour scheme as a whole, and allowed the program to generate a lot of different colouring possibilities - and to choose between them by controlling the ratios between the two sets of values for lightness and saturation. The result was, importantly, that AARON wasn’t following a human-type strategy. As Cohen put it: ‘it was almost as if, after years of requiring the program to do things my way – if such-and-such, do this – the program had finally said, just tell me what you want done and I'll do it my way.’ Silent Canyon 2 provides an example of the computer making such decisions. It’s a nice illustration of how the creativity lay in neither the programmer nor the program alone, but in the dialogue between the two – as built up between AARON and Cohen over several decades.


Cohen also sought to develop how AARON handled form. One challenge, emerging from a commission to make a portrait of his poet friend, Jerome Rothenberg, was to persuade AARON to produce a recognizable likeness. ‘As a result’, explained Cohen, ‘for almost two months I found myself developing AARON's knowledge of the structure of heads and faces. The number of data points involved grew from a few dozen to several hundred... The points were organized into parts — upper mouth, lower mouth, beard, forehead, eyelid, lower eye and so on — in such a way that the individual parts could be scaled and moved, three-dimensionally, at will[iii]’. Once he had a likeness, it would hold as the head turned and the facial expression changed. Along the way, moreover ‘AARON was generating make-believe people’ and ‘by the time the exercise was done, AARON's much-extended data base represented a prototype figure only, and the program had enough knowledge to generate from it a varied population of highly individualized physical and facial types, with a range of haircuts to match’ – including a representation of Jerry.

From the Suite for Discovery Channel Online, 1995. Dye on paper, 133 × 173 cm.

In the series Suite for Discovery Channel Online, 1995, the type and positioning of the painting arm allowed for further variation: they are pure computer paintings, but dripping the paint from a height to yield a blotchy effect suggesting the semi-accidental effects more typical of human application. Cohen also developed a new version of the program in which everything was built from an abundant supply of a single element: a small, hexagonal cell which attaches itself to the corners of other cells. Once a colony of cells has reached a certain size or complexity, AARON draws its enclosing boundary, and places this newly generated formal element into the developing image, as is visible in in[Ma1]  Another Spring (for R.C.).

Another Spring (for R.C.), 2011. Oil over pigment ink on canvas, triptych, 213 x 366 cm

Cohen’s next substantial shift, now that AARON was capable of developing both form and colour by stimulatingly unhuman means, was to question whether the results looked too ‘untouched-by-hand’: if an image shows no evidence of the manipulation of material, then it becomes that much harder to believe in its intentionality. So he started painting in oil over AARON’s ink images and their backgrounds, so ‘opening the door to the assumption of intentionality in the reading of the image.… Now I was contributing something the program had been unable to do.’ Cohen applied this insight to increasingly simplified shapes, moving away from the figuration which characterised AARON during the previous decade, and returning from 2012 towards the painterly abstraction he had been making the old-fashioned way in the mid-sixties. First Sighting is an example.

First Sighting, 2012. Oil over pigment ink on canvas, 122 × 220 cm.

Cohen never faltered in his drive to take his and AARON’s practice forward: in his last two years he experimented with adding his own interventions on top of the computer’s initial output by ‘painting’ with his fingers on a 55 inch screen, so integrating his manual interventions with the technology. ‘Once mixed’, he explained, ‘colours are stored in ‘pots’, and AARON records the identifying number of the pot, rather than the colour itself, for the path of each finger movement. That means that I can modify all the colours in a painting simply by modifying the contents of the pots, with none of the overpainting that would be required in a physical medium. By extension, I can also specify a particular area of the painting and remove, modify or replace every example of a particular colour falling within that area. In practice, it means that I think about colour more in terms of colour relationships than I had previously.[iv]’  Once he was satisfied, the file was ported to a wide format printer, where it was used to print the image on the canvas. The Last Machine Age is one of only 20 works he completed in this way.

The Last Machine Age, 2015. Pigment ink on canvas, 36 ×66 cm

What emerges, then, from almost 60 years of experiment at the interface of art and Artificial Intelligence is a substantial body of work which is compelling if judged in traditional aesthetic terms, but which also enables the viewer to track the narratives of innovation and collaboration which lie behind it – and to ask the question: who is the creator here?

For more on Harold Cohen, please visit the Gazelli Art House website.

The Gazelli Art House show Harold Cohen: The AARON Retrospective, is running 14 Oct – 19 Nov, 2022.

 
Notes:

[i] Harold Cohen: Driving the Creative Machine, 2010

[ii] Harold Cohen: AARON, Colourist: from Expert System to Expert, 2006

[iii] Harold Cohen: The Further Exploits of AARON, Painter, 1994

[iv] Harold Cohen: Fingerpainting for the 21st Century, 2016

Cohen in finger-painting action.

All images shown courtesy of Gazelli Art House. Photography Deniz Guzel. © Harold Cohen Estate, except Silent Canyon 2 and The Last Machine Age which feature in the Gazelli Art House show Harold Cohen: The AARON Retrospective, 14 Oct – 19 Nov, 2022.