The Drowned World project employs text-driven image synthesis to explore the ambiguous distinction the media introduces between imagination and virtuality. The machine learning technology in these explorations (VQGAN+CLIP) is an amalgamation of a generative adversarial network (VQGAN – Vector Quantized Generative Adversarial Network) and a natural language classification engine (CLIP – Contrastive Language–Image Pretraining). The CLIP component guides the GAN to try to produce images that match a text description. What emerges is a kind of language-mediated cybernetic Impressionism that appears to hold an intentionality relation between mind and artefact. The seeming endlessness of this experience may induce a kind of sub-aesthetic dopamine addiction. In The Drowned World project, I conduct visual investigations of this emerging space of the technological reinforcement of belief.
The liminal stability between belief and dream is the thematic ground of J. G. Ballard’s dystopian climate-warning novel The Drowned World, which I reference as a topical guide. I am interested in the possible worlds that emerge at the intersection of ways of knowing, at the discursive and non-discursive interstice. Drawing from a dream sequence found in Ballard’s novel, in the lefthand column above I gradually eliminate contextual referents in the text to generate the successively abstracted images on the right, progressing toward a literally depersonalized landscape devoid of life. The series speaks to the realities that emerge from our presumptuous preferential sampling of what technology has already determined to be “real.”
This image captures a triad of possible worlds in the text-image latent space resonant with my reading of the novel in terms of the digital Anthropocene. Semantic prompts and hand-painted cue images are merged and transformed as the AI optimizes the data in higher dimensional statistical space. Results are in turn reinput into the synthesis cycle, progressively perturbing and inspiring movement through the space of the virtual apparatus of artificial imagination. The text records Ballard’s reflections on dreams—these images then are a dream of a dream of a dream; dreams “all the way down” constituting of a sequence of poetic destructions and associations allowing for the mining of an affective aesthetic space implicitly encoded into the writing, drawing out the tacit image Ballard may have held in imagination at that time. The naming of the image is interesting to me, reflecting an unlikely autographic extension but simultaneously suggesting the senescence of computational relevance to the point of resonance. The series is part of a project on embodied aesthetics that suggests this polarity itself is an illusion.
These paintings were autographically painted in response to the novel’s primary dream sequence. The abstract expressionist style functions as initialization “noise” influencing network processing of the image sequences to follow. The three paragraphs of the novel’s dream sequence are iteratively combined with each of the three paintings to create a matrix mapping of latent space.
This image represents a mapping of AI mediated latent space in the text-image synthesis employed in The Drowned World project. I explore aesthetic affordance in such content matrices. Here, the semantic space of a dream described in J. G. Ballard’s novel is mapped onto autographic painting space. The grid represents a small selection of the exploratory sampling conducted; left-to-right are the 3 prompts from the novel and bottom-to-top major groupings are the resultant spaces generated by the three paintings on the left, a triptych hand-painted by the artist in response to themes in the novel.
Suk Kyoung Choi 