(A)rt man(I)festo
An exploration of AI and art: When human and computer both have roles in developing artworks,
whose creations are they? Who conceived them and who controlled them?
TOOLS
TensorFlow (textgenrnn, style transfer)
DeepDream
Adobe Illustrator
Collage
YEAR
2019
Concept
While taking A.rt I.ntel, I reflected on how the class had sparked my interest in the implications, hypotheses, and debates that surround artificial intelligence. The technical aspects of AI are certainly fascinating, but discussions about its present and future led me to think more critically about our relationship and coexistence, as humans, with this "other" kind of intelligence.
Earlier in the semester, my professor had shared an article titled "The AI-Art Gold Rush Is Here" (by Ian Bogost), which describes an exhibition of large prints generated by a machine-learning algorithm; the work is considered to be a “collaboration between an artificial intelligence named AICAN and its creator, Dr. Ahmed Elgammal”. The reading led to a conversation about the role of technology in art, the meaning and value of art itself, the question of authorship, and the definition of creativity and talent. Inspired by these points and questions, I decided to explore the intersection between AI and art.
The project has two parts: (1) an art manifesto inspired by human creations (existing art manifestos) but generated by artificial intelligence, and (2) a series of artworks that adhere to the manifesto, thus inspired by AI creation but generated by a human (me). I ended up creating five collages that correspond to five sections from the generated manifesto. The images that compose these collages were obtained by and transformed with algorithms also, but curated and arranged by me. And thus, having interwoven the algorithms' work with my own through all stages of the process, the question arises: Is it my work? Is it the algorithms'? Who conceived it, and who controlled it?
Manifesto
To create bodies of text, I used the Python module textgenrnn, built on Keras/Tensorflow. The first step to create the art manifesto was to find existing manifestos and artist statements to train the algorithm. The following are some websites and articles that were extremely useful as sources of text:
10 game-changing art manifestos / Top Five Artists’ Manifestos Of All Time / Manifestos / 8 Artist Statements We Love
Having compiled multiple manifestos and statements in a txt file, I uploaded it to the Interactive textgenrnn Demo w/ GPU (Collaboratory Notebook) and was able to obtain the new manifesto. In the image below, the manifesto is presented in a way that juxtaposes the "artificial" with the human, and beyond that, hints at the idea of imitation. The manifesto's text appears in a "programming" font, alluding to the machine origin of the text. The first sentence, "Space.", is shown as an image that has been processed with DeepDream. However, the page has several brush stroke elements, which commonly associated to non-digital work - yet these are shapes from the Adobe Illustrator software. Much like the text itself, the strokes pretend to be human-created, even though they are computer-generated.
With the new manifesto ready, I proceeded to select five sections I wanted to represent. These appear highlighted in different colors.
Collages
Once I had specific text to follow, my next step involved none other than Google's search algorithm. My strategy to find the images for the collages was to carry out Google searches with clusters of terms from each of the manifesto's sections. When the selection of Google images was ready, I used the deepdream.py and neural_style.py Tensorflow implementations to apply DeepDream and style transfer on some of my images. For style transfer specifically, I identified which images could work as content and which could work as style.
For example, for the fourth section ("...the mathematic and parts of the sofas, to live in freedom to the spirit of the state of the steel of the imagery of the stage..."), the following images show examples of how style transfer and DeepDream were applied.
In the final step, the manipulation and arrangement of the images was entirely up to me, giving me full agency in terms of the creative process. As was pointed out by my professor, though, despite transforming (by cutting and overlapping) and placing the elements of the collage in random and seemingly non-sensical ways, some patterns were ultimately followed that grant meaning to the elements. This is, ultimately, what the neural network in textgenrnn did: it followed a process that often produced random and incoherent results, including several nonexistent words, but which ultimately adhered quite impressively to the general patterns of the English language.
The final collages:
Exhibition
The following photographs show the layout of the manifesto and the collages during the NYUAD Interactive Media end-of-semester showcase, where they were exhibited alongside other students' work. To succinctly explain the project's intention and process to viewers, a small description was put up among the collages, also shown below.