What Are We Saying When We Call
an Algorithm ‘Creative’?

The artist Harold Cohen spent his life trying to create code that might be regarded as creative in its own right. Cohen began his career intending to be a conventional artist, and he seemed to be well on his way to achieving this goal when he represented Great Britain at the Venice Biennale in 1966, at the age of 38. Shortly after the show he met his first computer, thanks to a visiting professorship at the University of California–San Diego, where he met Jef Raskin. “I had no idea it would have anything to do with art,” he would later tell the Christian Science Monitor. He just got turned on by the programming aspect of the computer.

“It slowly dawned on me that I could use the machine to investigate some of the things that I thought I hadn’t been able to in painting, that had made me very discontented with my painting.” Raskin, who went on to create the Macintosh computer at Apple in the late 1970s, turned out to be a great choice of teacher. (The name was chosen because McIntosh was his favorite variety of apple; the spelling had to be changed for legal reasons.)

Inspired by Raskin, Cohen went on to produce AARON, a program he wrote to make works of art. Cohen’s code was of the top-down, if-then variety. By the time he died, in 2016, it consisted of tens of thousands of lines. What was interesting to me was how Cohen described the code’s creation process. He talked about AARON “making decisions.” But how had he programmed these decisions?

People involved in creating computer art tend to be reluctant to reveal the exact details of how their algorithms work. This subterfuge is partly driven by their goal of creating algorithms that can’t easily be reverse-engineered. It took some digging in the code for me to find out that “making decisions” was code for Cohen’s choice to put a random number generator at the heart of the decision-making process. Like Nees, Cohen had tapped into the potential of randomness to create a sense of autonomy or agency in the machine.

Is randomness the same as creativity? Many artists find that a chance occurrence can be a helpful spur to creation. In his Treatise on Painting, Leonardo Da Vinci described how a dirty cloth thrown at a blank canvas might serve as a catalyst for the next step. More recently, Jackson Pollock allowed the swing of his bucket to determine his compositions. Composers have found that chance sometimes helps them head in new and unexpected directions in their musical composition.

But randomness has its limitations. There is no deliberation going into a choice of one configuration as more interesting than any other. Ultimately it is a human decision to discard some of the output as less worth keeping. Randomness is, of course, crucial when it comes to giving a program the illusion of agency, but it is not enough. It is still up to human hands to press the “on” buttons. At some point will algorithmic activity take over and human involvement disappear? Our fingerprints will always be there, but our contribution may at some point be considered to be much like the DNA we inherit from our parents. Our parents do not exercise creativity through us, even if they are responsible for our creation.

But is randomness enough to shift responsibility from the programmer to the program? Cohen died at the age of 87. AARON, however, continues to paint. Has Cohen managed to extend his creative life by downloading his ideas into the program he created? Or has AARON become an autonomous creative artist now that Cohen is no longer working as partner of his creation? If someone else now presses the “create” button, who is the artist?

Cohen said he felt his bond with AARON was similar to the relationship between Renaissance painters and their studio assistants. Consider modern-day studios like those of Anish Kapoor and Damien Hirst, where many people are employed to execute their artistic visions. At an old dairy factory in south London, Kapoor has a big team helping him, just as Michelangelo and Leonardo did.

Cohen was part of a whole movement of artists in the 1950s and 60s who started exploring how emerging technology might unleash new creative ideas in the visual arts. The Institute of Contemporary Arts in London held an influential exhibition in 1968 called Cybernetic Serendipity that profiled the impact that the robotic movement was having in the art world. It included Nicolas Schöffer’s CYSP 1, a spatial structure whose movements are controlled by an electronic brain created by the Dutch company Philips. Jean Tinguely supplied two of his kinetic painting machines, which he called Métamatics. Gordon Pask created a system of five mobiles that interacted with each other based on the sound and light each emitted. The interactions were controlled by algorithms that Pask had written. The audience could also interact with the mobiles by using flashlights.

At the same time, Korean artist Nam June Paik was building his Robot K-456, billed as history’s first nonhuman action artist. The original intent was for it to give impromptu street performances. As Paik recounted, “I imagined it would meet people on the street and give them a split-second surprise, like a sudden show.” As technology has grown ever more sophisticated, so has the art exploiting that technology. But how far can these robots and algorithms go? Can they really become the creators rather than the creations?

Simon Colton has been working on a program to take on AARON’s mantle. Here is what the Painting Fool, his creation, says about itself on its website:

I’m The Painting Fool: a computer program, and an aspiring painter. The aim of this project is for me to be taken seriously—one day—as a creative artist in my own right. I have been built to exhibit behaviors that might be deemed as skillful, appreciative and imaginative.

Of course, these are really the aspirations of Colton, its creator, rather than of the algorithm itself, but the aim is clear: to be considered a creative artist in its own right. Colton is not looking to use algorithms as a tool for human creativity so much as to move creativity into the machine. The Painting Fool is an ongoing and evolving algorithm that currently has over 200,000 lines of Java code running its creations.

One of Colton’s early projects was to create an algorithm that would produce portraits of people who visited the gallery. The results were then displayed on the walls of the gallery in an exhibition he called You Can’t Know My Mind. The portraits needed to be more than just photographs of visitors taken by a digital camera. A portrait is a painting that captures something of the internal worlds of both the artist and the sitter. But because the artist in this case was an algorithm without an internal world, Colton decided to algorithmically produce one. It needed to express (if not feel) some emotional state or mood.

Colton didn’t want to resort to random number generators to choose a mood, as that seemed meaningless. And yet he needed a certain element of unpredictability. To set his algorithm’s emotional state on any given day, he decided to have it scan articles in that day’s Guardian. I can attest that my own morning perusal of the newspaper can lift or dash my spirits. Reading about Arsenal’s 4–2 loss to Nottingham Forest in the third round of the 2018 FA Cup certainly put me in a foul mood—my family knows to avoid me when this kind of thing happens—whereas a preview of the final season of Game of Thrones might fill me with excited expectation.

The programmers would not be able to predict the state of the algorithm, as they wouldn’t know which article was influencing it when it was prompted to paint. Yet there would be a rationale as to why the Painting Fool chose to paint in a certain style.

When a visitor sat down for a portrait, the algorithm scanned an article for words and phrases that might capture the mood of the piece. An article about a suicide bombing in Syria or Kabul would set the scene for a serious and dark portrait. Colton calls the choice “accountably unpredictable.” The painting style isn’t simply a random choice—the decision can be accounted for—but it is hard to predict.

Sometimes the Painting Fool would be exposed to such depressing reading that it would send visitors away, declaring that it was not in the mood to paint. But before they left, it would explain its decision, providing the key phrase from the article it had read that had sent it into such a funk. It would also stress that “No random numbers were used in coming to this decision.”

This ability to articulate its decisions, Colton believes, is an important component of the dialogue between artist and viewer. In the exhibition, each portrait comes with a commentary which seeks to articulate the internal world of the algorithm and to analyze how successful the algorithm thinks the output is in rendering its aims. These are two components Cohen said he missed in AARON.

I asked Colton if he believed that the creativity of this activity came from him, or how much creativity he attributed to the algorithm. He very honestly gave the Painting Fool a 10 percent stake in what was being produced. His aim is to change the balance over time. He proposed a litmus test to this end, suggesting it would be “when The Painting Fool starts producing meaningful and thought-provoking artworks that other people like, but we—as authors of the software—do not like. In such circumstances, it will be difficult to argue that the software is merely an extension of ourselves.”

One of the problems Colton sees in mixing computer science and creative arts is that computer science thrives on an ethos of problem-solving. Build an algorithm to beat the best player of Go. Create a program to search the internet for the most relevant websites. Match people up with their perfect partners. But creating art is not a problem-solving activity. “We don’t ‘solve the problem’ of writing a sonata, or painting a picture, or penning a poem. Rather, we keep in mind the whole picture throughout, and while we surely solve problems along the way, problem-solving is not our goal.”

Cohen said more about the difference: “In these other areas, the point of the exercise is to write software to think for us. In Computational Creativity research, however, the point of the exercise is to write software to make people think more. This helps in the argument against people who are worried about automation encroaching on intellectual life: in fact, in our version of an AI-enhanced future, our software might force us to think more rather than less.”

The strategy of the team is to keep addressing the challenges offered by critics for why they think the output isn’t creative, beating the critics finally into submission. As Colton puts it, “It is our hope that one day people will have to admit that The Painting Fool is creative because they can no longer think of a good reason why it is not.”

AARON and the Painting Fool are both rather old-school in their approach to creating art by machine. Their algorithms consist of thousands of lines of code written in the classic, top-down mode of programming. But what new artistic creations might be unleashed by the new, bottom-up style of programming? Could algorithms learn from the art of the past and push creativity to new horizons?

__________________________________

Adapted from The Creativity Code: Art and Innovation in the Age of AI by Marcus du Sautoy, published by Harvard University Press. Copyright © 2019 by Marcus du Sautoy. Used by permission. All rights reserved.