“Autonomous” Computer Art
Is such a thing possible?
By Stuart Smith
Is a work of art conceived and executed by a computer without any human intervention possible? That is, is “autonomous” computer art possible?
A thought experiment
In an attempt to answer this question, I propose a thought experiment based on the infinite monkey theorem. This theorem states that a monkey hitting keys at random on a typewriter keyboard for an infinite amount of time will almost surely type any given text, including the complete works of William Shakespeare. In my thought experiment, the analog of the infinite monkey is a computer executing a random sequence of its own instructions for an infinite amount of time.
The setup
For this thought experiment I’ll assume that an area of the computer’s memory is designated as a sort of artist’s canvas. The canvas will be represented on the computer’s screen as a square grid of cells that can be either black or white. Whenever an instruction changes one of the cells from black to white or white to black, the computer will immediately update the screen. This process will continue indefinitely. Meanwhile, some infinitely patient human will have to watch the screen to see if anything that looks like art appears; without the human observer, we’d have to assume that the computer, asserting its complete autonomy, could say “it’s art if I say it is.”
Understanding the challenge
To get a rough idea of how likely the proposed process is to produce art, consider Figure 2. This is a stylized heart in a 64×64 grid of square cells. This is an extremely small grid for today’s computer graphics, but the number of possible black and white images in this grid is so large that when I asked my software to calculate this number it essentially threw up its hands and said the number is “Inf” (infinity).
Intuitively, “Inf” means that it could require eons just to produce a crude picture of a single recognizable object. But let’s not give up just yet. No, let’s look at some examples of what might happen if we ran some actual experiments.
Experiment 1: random images
The first demonstration shows the results when the computer is given “complete” autonomy, that is, when it is allowed to simply execute random instructions. Figure 3 shows a typical random cell image:
What we see here is what we would expect to be the result of the computer executing a completely random series of instructions. There is no discernible pattern. It’s just visual noise.
In fact, complete autonomy is impossible even at this level: a human will have to write at least some minimal program to get the process started and a human will also have to initiate execution of the program.
Experiment 2: looping
A less probable outcome would be a sequence of random instructions containing an instruction that jumps back to the beginning of the sequence, causing a single image or short sequence of images like Figure 3 to be generated over and over. There would certainly be a discernible pattern, but the repetition would quickly become tiresome.
Experiment 3a and 3b: cellular automata
An even less probable outcome would be a random sequence of instructions that form a “cellular automaton.” In this case a single sequence of instructions would be executed over and over, as in Experiment 2. But each successive image in the display would be derived from the previous image according to definite rules. These rules would have had to appear by chance in the random sequence of instructions.
a) Conway’s “Game of Life”
John Conway’s Game of Life is perhaps the most famous cellular automaton. The Game of Life begins with a random image like Figure 3, but it has three simple rules for deriving each subsequent image from its predecessor. Each rule says what happens to a cell in the grid based on the states of its eight neighbors (Figure 4). A white cell is “live” and a black cell is “dead.” Figure 4 shows the “neighborhood” of cell C, whose color will be determined by the colors of its eight neighbors.
To derive a new image from the current one apply these rules:
1. Any live cell with two or three live neighbors remains live.
2. Any dead cell with three live neighbors becomes a live cell.
3. All other live cells die in the next generation. Similarly, all other dead cells stay dead.
To fully appreciate the Game of Life you need to see it in action and, ideally, play with it. At the following website you can play the Game:
There are controls to start and stop the game, run it continuously or one step at a time, and to regulate its speed. You can select different starting configurations from a “lexicon” to see how the game evolves step-by-step from different initial arrangements of white cells.
Since the Game of Life was published in 1970, researchers and Game aficionados have developed ways to create interesting patterns by setting up the appropriate initial image. Of course, setting up the initial image completely deprives the computer of autonomy because once the initial image is set, the subsequent sequence of frames is generated completely deterministically.
b) Wolfram’s Simple Cellular Automaton
Stephen Wolfram published a “simple” cellular automaton that has gained almost as much notoriety and spawned as much research as the Game of Life. Wolfram’s cellular automaton operates according to a single rule (chosen from a set of 256 rules) which it repeatedly applies. The way this cellular automaton is usually presented, we see it “evolve” one line at a time. Each successive row is obtained by applying the rule to the preceding row. Figure 5 shows examples of the figures produced by particular rules.
Figure 5 is excerpted from https://mathworld.wolfram.com/ElementaryCellularAutomaton.html, which shows figures for all 256 rules and provides an explanation of how they work.
There’s clearly a lot of structure in these images. To get these results, the first line of the image must be a single black cell surrounded on either side by an equal number of white cells. As the computation proceeds, the sequence of instructions applies the rule to each cell and its left and right neighbors to get the correct value, black or white, to put in each cell on the next line down.
It’s easy to write a very compact program to implement the simple cellular automaton. But, like the Game of Life, this program necessarily has a very specific logical structure. It seems intuitively obvious that the exact sequence of instructions required would be highly unlikely to appear by chance.
The original presentation of the simple cellular automaton is in Wolfram’s book, A New Kind of Science (Wolfram Media, 2002). While the book’s ideas have found wide application in the natural sciences, mathematics, and economics, it has also had a surprisingly strong impact on digital art. The book is filled with attractive figures depicting various computational processes, and completely apart from its technical value it stands out as a visually striking “coffee-table” book on a par with volumes on the works of great visual artists. The book is unquestionably a work of art in addition to its other attributes.
Experiment 4: a complex cellular automaton
The following cellular automaton generates a sequence of patterns from an initial colored image.
The dynamic behavior of the complex cellular automaton can be seen at:
Run the automaton and observe its behavior. As cells “consume” their neighbors and get within range to be consumed by higher-ranking cells, the automaton goes into a consuming phase, where blocks of color advance against the remaining blocks of randomness. At this point, objects called “demons” appear. Demons continuously rotate and generate waves that spread out in a spiral pattern. Smaller demons consume larger demons until every cell of the automaton eventually enters a repeating cycle. The video ends at the point where this occurs.
This cellular automaton has clearly been programmed to go through a specific complex sequence of actions. The result has an artistic aspect, but this is due to the skill and imagination of the person who created the automaton.
AI-generated art
Some people believe that autonomous computer art can be produced by Artificial Intelligence. As is well known, programs such as Stable Diffusion can create images given only a verbal description (i.e., a “prompt”) of the desired image. The picture at the top of this article, for example, was generated by Stable Diffusion Online from the prompt “a monkey typing on a computer keyboard.”
Stable Diffusion is a sophisticated program that has been “trained” on an enormous array of paintings, drawings, photographs, and other kinds of images. It is also given information about each image: who created it, when it was created, what it depicts, etc. When it receives a prompt, it uses all this information to create an image that meets the requirements of the prompt. In this respect it works as a kind of “auto-complete” in that it creates the most likely picture given the information in the prompt. In no sense is the computer acting “autonomously.” Until the computer begins to generate its own prompts without human input, this process cannot be said to create autonomous art.
