Turing Test for Art: A. Ramachandran Case Study
by Rahul Ramachandran
Like everyone during the pandemic, I was stuck inside four walls. Needing to expend my energy and distract my mind from all the negative and depressing news about divisive politics, the spread of COVID, and the general deterioration of human society, I started exploring new AI algorithms such as generative adversarial networks (GANs) to create art. These algorithms take two inputs — the first is a base image, which is typically your photograph, while the second is a style seed image, typically of a well-known work of art such as Van Gogh’s The Starry Night. The algorithm then iterates to change your photo and style it like Van Gogh’s work. The results are mostly kitschy yet good for amusement.
When I started experimenting with these algorithms, I was more interested to see how I could induce randomness in the process so that the result did not look like a weak attempt at copying a style of a master but at creating a new style pattern and texture that was unique and interesting. An example is my Hallucination series which uses my photographs from Antelope Canyon in Arizona as the base image and uses an Islamic art tile as a style seed to create tessellation patterns. Both the source image and the final Hallucination series image are shown below.
Both my parents are well-known artists (A. Ramachandran and Chameli Ramachandran). I have always wanted to collaborate with my father on some creative project but never knew specifically what that could be. My collaborations with him growing up were always limited to helping him with stretching canvas and moving his large paintings from one place to another to show visitors to his studio. Later in life, I created and still maintain his website. While his website is meant as a medium to reach his fans worldwide directly and as a knowledge repository about his lifetime of creative endeavors, creating and regularly updating his website has allowed me to stay connected with his works even though we now live thousands of miles apart.
On my last visit to see my parents, I saw all the new paintings my father completed during the pandemic. Focusing on his creative process was his coping mechanism during the pandemic. I also reconnected with an old schoolmate, and she invited me to explore places in Delhi that I had not visited for over two decades. Based on these two experiences, the concept of Snapshots of a Moment in Time germinated in my mind. In this series, I wanted to use visual imagery that was very Indian and an inherent part of my father’s paintings. I also wanted to stylize it not based on his paintings but on a motif within his paintings — how he visually represents clouds or hills.
Once I completed the series, I was curious to treat this imagery set as a “Turing test for art,” specifically my father’s art. What did he think about the results? What did I think of the whole experiment? I would like to add a caveat that, in many ways, this is not a true Turing test. To generate my neural art, I use the AI algorithm as just one step in the creative process. The AI algorithm is primarily used as an augmentation tool for stylization.
I shared my result — a full set of works with my parents to review and critique objectively. And I was not spared from criticism.
Compositions are not good.
Figures need to be more prominent; figuration needs to be better.
Some compositions are too sentimental, and there is too much realism in others.
Tonal variation in the colors while stylization is not correct.
Tum ko color sense nahi hai. (You don’t have any sense of color.) [I am color blind.]
Out of my set of thirty-two AI-augmented neural art, they finally selected eight images.
“Yeh theek hai” (These are ok). When I explained the methodology and the algorithm, my father’s response was a classic A. Ramachandran retort: “Interesting, but too complicated; painting is better.” What was my parent’s final verdict? “Acha hai (Good) experiment. Keep trying.” He reiterated that he would like to have been involved in the process from the beginning. He would have liked to help with the initial base images, especially with their composition. He recommended that I use less complicated images and fewer colors in the future.
“Can this be done better by someone with more artistic sense?” I asked my father.
Yes, but it needs to go beyond ordinary painting concepts. One could combine two or three different paintings to create a whole new kind of imagery — only then be interesting and original. Some of the compositions and motifs were too obvious and too close to the source material [his works] to be interesting and original.
You can create quite fantastic art images using this methodology, but you need to ensure that the base image is not ordinary. The final art image generated needs to be innovative. Play with the juxtaposition of images while composing the base image. In the end, it does not matter what technique you use, a paintbrush or a computer algorithm, the art itself should be striking, interesting, and not ordinary.
My main takeaway from this experiment is that AI algorithms are ultimately still tools and are not replacing the artists. Most of these algorithms attempt to replicate styles, colors, and patterns, not create anything new or original. To get something new, unique, and truly innovative, one has to start with an interesting original visual idea and then use the algorithm to induce randomness into the stylization. On a personal note, I would love to get a chance to collaborate with my father on one image from start to finish. This will be difficult given his age and the limited time we get to be together. While the opportunity to learn and create something with him would be amazing, the memory of the experience itself would be far more precious.
