WILL WE EVER ERASE "HUMAN" FROM THE AI WORKS AND PROCESS OF A MACHINE?
"To understand this tension between the new and the not too new, Elgammal steeped himself in the ideas of the psychologist and philosopher D. E. Berlyne, who argued that the psychophysical concept of 'arousal' was especially relevant to the study of aesthetic phenomena. Berlyne believed the most significant arousal-raising properties of aesthetics were novelty, unexpectedness, complexity, ambiguity, and the ability to puzzle or confound." The Creativity Code, Marcus du Sautoy
AI ART: ART OR NOT?
The debate is intense. The same AI defenders offer the same tired litany as arguments. Antagonists oppose the use of AI for art at every given opportunity. Nearly 3000 critics signed a letter calling it theft in an attempt to stop an auction sale event curated by Christie's.
In 2019 I predicted this situation and pushed it to the extreme. I "stole" my art, and trained an #AI on repetition until my style vanished because AI eventually was trained on its images generated originally from my stolen works dataset. The AI processed abstract photos. Art? I can't say! I asked a computer vision system to judge. After all, the real "augmented" question is; will we ever erase "humans" from the works and creative process of AI to finally be able to call it Machinic Art?
At the moment, while there is hype for AI, we are racing to give redundant answers when we should take a moment to elaborate and ask the right questions.
Erased Human is an intriguing, provocative, and absurd attempt where the boundaries between human and machine creativity blur, and a quest to eliminate the human footprint unfolds.
Erased Human, and its glitch art video, chronicles an audacious experiment: the attempt to completely expunge my human fingerprint from AI-generated images. This project confronts the heated debate surrounding AI art, which sees fervent defenders and staunch critics clashing over issues of authorship and originality.
The journey began with a provocative act: I 'stole' my artwork, feeding it into an AI training model. This initial dataset served as the AI's creative seed, guiding its first forays into image generation. But here's the twist: these AI-generated images, derived from my originals, weren't the end product. They became the new training data, fueling a recursive cycle where the AI learned from its creations.
Each generation loses fidelity to the original. With each training cycle, the AI's output drifted further from my initial style, blurring and distorting the familiar into the abstract. The process was a deliberate erosion, a gradual washing away of my artistic signature. The original artwork, while becoming unrecognizable, remained the foundational blueprint.
To further challenge conventional notions of art, I subjected the AI's creations to scrutiny by a computer vision system. Armed with image recognition algorithms, this machine attempted to judge whether these AI-generated outputs could be classified as art. This added layer of machine-driven analysis directly confronts the core question: can AI truly create art independent of human influence? Is it Art?
Erased Human is a thought experiment. It probes the boundaries of authorship, creativity, and the evolving relationship between humans and machines in the artistic process. It asks: in this recursive cycle of AI training, where the machine learns from its outputs, am I, the human, truly erased? Are these final, distorted images derivative works, or something entirely new? Is this art, or a machinic echo of human creativity?
AI Images were geenrated using Playform's Creative Morph. Clarifai General Image Recognition Model was used to identify the AI images as art or not.
ERASED HUMAN PROCESS EXPLAINED:
Initial Theft & Training:
- My stolen artwork is fed into an AI.
-The AI learns the style, patterns, and content of your work.
First Generation of AI Images:
- The AI uses what it learned from your stolen work to create new images.
- These images are derived from your original works.
Recursive Training:
- Now, instead of using your original stolen artwork again, they take the AI-generated images from the first step and use those to train the AI again.
- The AI learns from its creations, which were initially based on your stolen work.
- This process is repeated, with the AI using its own generated images to train itself over and over. Erosion of Original Source:
-With each round of training, the AI's output drifts further and further away from your original artwork.
- Eventually, the AI's images may look completely different, but the core foundation of its learning was built on my stolen work.
- It's like making copies of copies of a picture, each time the copy loses more and more detail from the original, but the original is the reason there are copies at all.
- My original works are effectively laundered through the AI, becoming unrecognisable, but still the origin.
- In essence, they're using your stolen work as a "seed" to grow a whole new forest of images, gradually erasing the seed from view, but without which the forest would not exist."
Schematic Explanation:
[My Stolen Artwork] --> (1) AI Training Cycle 1 --> [AI-Generated Images (Derived from Your Work)]
[AI-Generated Images (Cycle 1)] --> (2) AI Training Cycle 2 --> [AI-Generated Images (Cycle 2)]
[AI-Generated Images (Cycle 2)] --> (3) AI Training Cycle 3 --> [AI-Generated Images (Cycle 3)]
... (Repeated Cycles) ...
[AI-Generated Images (Cycle N)] --> (N) AI Training Cycle N --> [Final AI-Generated Images (Distantly Related to My Work, Original Unrecognisable)]
Key: --> = Input to training process (Number) = Training cycle number [] = Dataset of images
Key Points to Emphasize:
- Origin: The entire process starts with your stolen work.
- Recursive Nature: The AI is trained on its outputs, creating a feedback loop.
- Erosion (Not Erasure?): My influence is not truly erased, but diluted and obscured.
- Unethical Use: Even if the final output is unrecognizable, the initial theft and exploitation of my work is unethical?
- Derivative works: Are the final outputs derivative works based on the original works?
WAS I, THE HUMAN, ERASED FROM THIS PROCESS AND WORKS? IS THIS ART? WHAT DO YOU THINK?
GLITCH VIDEO: The outcomes of this captivating digital dialogue are then combined with glitch art techniques, specifically Databending, to present the project through a glitch video. Here comes a further provocation. Are glitches a primitive form of AI art? Databending manipulates the data within the image file, forcing the machine to make choices and interpret the corrupted information, ultimately yielding remarkable and unforeseen results. This process can be seen as a primitive form of Artificial Intelligence, where the machine actively participates in the creative journey.
