Art and AI
I have experimented with Artificial Intelligence in my practice at different points over the years, most recently using Midjourney to generate reference images and to explore how different ideas might be expressed visually. I’ve used chatGPT to generate text in the same vein - exploring the expression of an idea that I was trying to clarify and give birth to. Recently, I stumbled across theobot.ai. It promoted itself as “the perfect writing assistant for artists”. If you have ever stared at a blank page trying to conjure a meaningful statement about your work, you can immediately understand how that promise grabbed my attention. A free five day trial was offered, so having nothing to lose, I jumped in to try it out.
I selected six images from my recent mixed media work as my source material (pictured below). Three images (bottom row) were from a discrete body of work that I have been developing that is focused on the sisterhood of women over time and space in our roles as givers of life, healers, storytellers, and carriers of culture, and what it means to be mothers and daughters, sisters and friends in a great unbroken chain, standing as a link between those behind you and those who will carry on after you are gone. I then added three images (top row) that are only loosely related, if at all, to the first three images. I included images that I considered successful and images that I considered less so. The diversity was intentional - I wanted to see what the bot made of it all.
The bot web site was straightforward and easy to use. There are options to upload images, to analyze individual images, to analyze sets of images, to generate titles, to generate web site content, to generate promotions for events, etc. In my first step, I uploaded the six images I had selected and asked the bot to generate descriptions of each in turn. At this point, the system had no other information than the individual image itself. I was pleasantly surprised in its accuracy in “seeing” and interpreting each work. It figured out that the works involved printing and collage on paper, though it missed some obvious symbols (e.g. the universal icon for female), and noticed only two background figures in the bottom left image rather than the three that are there. It also interpreted some things in ways that I had not intended. Where I use circles as symbols of unity and connection, it saw eyes. I accepted these other viewpoints as potentially useful, and cataloged them as fodder for additional thought.
The second step of the process allowed me to add my own information to each uploaded image - a title and description, as well as information on the process and mediums and the inspiration behind the work. After entering this data, I generated new, more specific descriptions of each image. These obviously played on my own words, but added depth and occasionally a new perspective. I should probably take an aside here to admit that I can hardly stomach much of the “art speak” I read and hear. (If I hear the word liminal again in my lifetime, it will be too soon.) The generated text did not trip my gag reflex, though it was more sugary and self congratulatory than I’d use. The system offers a choice of tone, so I cycled through each option, hoping to eliminate some of the hyperbole. I found useful phrases and ideas in each version, though each continued to strike me as a bit over the top.
Let’s look at an example using the “Mystical and Enigmatic” tone, using the image below.
The description generated was:
While the notions of mystical and enigmatic felt appropriate for the nature of the work and themes I was exploring, the text is a bit breathless for my tastes. Though admittedly, if I toned it down just a bit, I could see the description having value, particularly in a marketing context.
Having completed the descriptive information for each image, I then asked the system for a critique of each work. There were good nuggets of feedback in the results, the system confirming in several instances the weaknesses that I myself knew to be in the work, and noting strengths that I felt were valid. But again, the text was often overblown . The bot seemed to be as concerned about protecting my self esteem as it was about giving me constructive and critical input. If I ask for a critique, I want honest feedback, good, bad, and indifferent - all the congratulatory language just made me doubt the validity of the critique. Let’s look at the critique of the same image.
It will take more exploration and analysis to detect the real value here - is it a useful voice for those of us with limited access to art world mentors or is it simply wrapping our own words in dopamine laced compliments and a few fairly obvious suggestions? Time and deeper, more considered use will tell.
These first 3 steps - 1)upload and generate a raw description, 2)add my own words and generate an informed description, and 3)ask for a critique - were taken on each individual image in turn. I then had the option to analyze a set of images for consistency. Sadly, this option is limited to four images. Allowing ten or fifteen images to be analyzed would be more useful in assessing how well a body of work holds together. When I submitted some of the series images with an unrelated image, the result was unexpectedly a bit mealy mouthed. When I chose the three series images the system generated the following:
In terms of value in helping me to better assess and better articulate my work, I found this aspect of the system least helpful. There was nothing in the consistency analysis that challenged me or gave me new perspectives to consider.
In the final step of my evaluation, I asked the bot to generate an artist’s statement. It follows.
I have very ambivalent feelings about the impact of AI on our society. I’ve experienced the Social Dilemma first hand from inside the industry and I fear the dark side of AI technology may greatly outweigh the good. In the end, however, I found the system interesting and useful enough to want to play with it further. I signed up for a monthly subscription at the rate of $5/month. For the price of a cup of coffee, I’ll give myself time to probe it further.
As it stands, the system provided a sense of having immediate access to informed viewpoints other than my own, something that we all need and often have little access to. And while I don’t expect to ever use any of the generated text verbatim, it did open my eyes to useful perspectives and points of view, and it provided some usable turns of phrase that I am unlikely to have come up with on my own. Another aspect of the experience encouraging me to continue my exploration was the responsiveness of the team behind the tool. I completed an evaluation giving my critical feedback to them, and I contacted them via email regarding language in their Terms and Conditions that I questioned. In both cases, I got prompt, friendly, and helpful acknowledgements.
As I go forward, I will put more thought and intention into the text I feed the system, explore all the options provided more fully, and apply the system to a broader range of my work. I will occasionally report back, particularly if something noteworthy emerges. If you choose to give it a whirl, I’d love to hear about your experience.
Images, text, and video by Wanda Oliver, all rights reserved.