Can Computer Art be Creative?

A Paradox

Figure 1. Image by the artist

Recently I had my app for creating digital art run off a series of 100 images. The image above was number 86. I didn’t know exactly what kinds of images to expect since most of the image-generating routines in my app take random inputs that determine what the images will look like.

While looking over the results I got, I had a recurring thought: where is the creativity in this? The computer simply did what my app’s program told it to do. And because key inputs to the program were random, you couldn’t say that the program was being creative. Perhaps the choices I made in designing my app constituted the creative act? What is creativity anyway?

An expert view of creativity

As I was thinking about this, I was reminded of a passage in a paper written over 30 years ago by the late William J. Mitchell, Dean of the School of Architecture and Planning at the Massachusetts Institute of Technology. This paper was mainly addressed to architects and to designers in related fields, but I think it applies to any human activity where creativity is believed to be essential to great work. Certainly all the arts are in this category.

I’ll quote the passage from Mitchell in full since I believe it gets at the heart of the matter.

‘Creative’ design appears to be a residual category: it encompasses all the things that designers do for which we cannot specify an effective and efficient mechanism. This presents a paradox. Any successful attempt to describe the mechanics of some ‘creative’ design activity will have the immediate effect of redefining that activity as ‘noncreative’. The more success we have, the more we can be accused of dealing only with the noncreative aspects of design. Undaunted by this, my aim here is to reduce the residue — not to nothing, but to something rather smaller than it is usually taken to be. (William J. Mitchell. A Computational View of Design Creativity, in John S. Gero and Mary Lou Maher, eds. Modelling Creativity and Knowledge-based Design. Lawrence Erlbaum Associates, Inc., 1993).

Some people now apparently believe that AI programs like Chat-GPT and Stable Diffusion have already reduced the residue to zero in literature and visual art, leaving no role for humans to play.

What if I reduce the ‘residue’ to zero?

According to Mitchell’s logic, I could imagine the following scenario for my own work: as I have my app generate more and more images, I find that I can continually narrow the range of the random inputs to the app until I finally converge on precisely one specific value for each input. The set of these values together yields a “perfect” image of a particular type. As a result, whenever I run my app again, it will simply crank out identical copies of that perfect image. I haven’t arrived at this point yet. If I had, it would certainly have taken all the fun out of making digital art.

A few technical remarks

It takes my app about 6 seconds on average to generate a square, 512×512-pixel image on my home system. Each time the size of the image square is doubled, the time per image goes up by a factor of 4. The largest image the app can generate in a tolerable amount of time is 2048×2048 pixels, which when printed is about 21 inches square. The images are all square because some of the app’s most important image-processing routines require square images as input.

A “normal” run of the app produces some number of variations on a single image template. It typically takes around 32 images to produce one that is worth saving and exhibiting. For this article the app was set up to randomly generate single images from 32 different templates.

I occasionally apply a post-processing step to my app’s images with GIMP, which provides a wide selection of filters to pretty up an image; however, all the images here are shown just as they were delivered by the app.

Selections from the run of 100

What follows is a selection of eight images from the run of 100 that led to this meditation on creativity. To me these are the best ones from that collection, but not necessarily the best my app has ever produced. Even though the generation of the images was automated, the selection of images for presentation here was still in human hands.

Figure 2. Image by the artist

Figure 3. Image by the artist

Figure 4. Image by the artist

Figure 5. Image by the artist

Figure 6. Image by the artist

Figure 7. Image by the artist

Figure 8. Image by the artist

Figure 9. Image by the artist

Copyright © Stuart Smith, 2023. All rights reserved.