AI
Generative Adversarial Networks (GANs) are inherently an asemic practice because they separate the form of writing or symbols from their semantic meaning. [1]
Asemic writing is a wordless, open semantic form of art that mimics the visual structure of writing without containing actual verbal content. When GANs are trained on textual, calligraphic, or typographic datasets, they do not read, comprehend, or intend to communicate linguistic truth. Instead, they map structural and mathematical distributions of pixels, strokes, and negative spaces. [1, 2, 3, 4, 5]
The resulting outputs look undeniably like text, symbols, or ancient scripts, yet they remain entirely unreadable. GANs achieve this unique form of digital asemic practice through several distinct mechanisms: [1, 2, 3]
1. Form Without Content (Pure Mimicry)
The core of asemic art is the illusion of writing. GANs excel at this because the Generator learns to replicate the statistical distribution of lines, curves, loops, and alignments found in human handwriting or typography. Because the machine lacks consciousness, human culture, or an understanding of grammar, it can only replicate the gestalt (the visual essence) of text. The output possesses textual "body language" but contains no underlying code, vocabulary, or message. [1, 2, 3, 4, 5]
2. The Adversarial "Fakery" Loop
In a GAN, the Generator constantly tries to fool the Discriminator into believing its creations are authentic. [1]
- To win this mathematical game, the Generator strips away the requirement for actual meaning and focuses entirely on believability of form. [1]
- It discovers the exact threshold where a cluster of pixels "looks" like a letter to the Discriminator without actually being a letter. [1]
- This mimics the human asemic practice of doodling or painting pseudo-scripts that trick our brains into trying to decode them, only to find a vacuum of meaning.
3. Latent Space as a Field of Infinite Variations [1]
When an artist interacts with a GAN's latent space (the hidden, multi-dimensional vector space where the AI maps features), they can smoothly navigate between different typographic styles or characters. Traivailing through latent space produces fluid morphing, warping, and hybrid symbols that sit precisely in the uncanny valley between known written scripts. It creates transient, fluid glyphs that belong to no human language but look as if they should. [1, 2, 3, 4, 5]
4. Glitching and Artifacts as Gesture
Traditional asemic writing relies heavily on the physical gesture of the hand—scratching, flowing, and smudging ink. In a GAN, this human gesture is replaced by algorithmic gesture. During early training phases or during a "mode collapse," GANs produce abstract, bleeding, and fragmented letterforms. These digital artifacts function exactly like painterly ink splatters or illegible medieval script fragments, highlighting the machine's internal processing as an expressive medium. [1, 2]
5. Shifting the Burden of Meaning to the Viewer [1]
Asemic art is defined by an open semantic interpretation; because the text has no fixed definition, the viewer must project their own meaning onto it. GAN-generated text does the exact same thing. When looking at a GAN's unreadable, synthesized handwriting, the human brain naturally tries to find familiar shapes—graphemes, words, or symbols. The viewer becomes the writer, filling the AI's void of communication with their own narrative. [1, 2]
If you are researching this intersection of machine learning and art, we can explore:
- Specific AI art projects utilizing textual GANs
- The difference between how GANs and Large Language Models (LLMs) treat text
- How to use latent space manipulation to control the "readability" of generative art
No comments:
Post a Comment