A.I. and Hauntology
Artists using A.I. image generation often consider themselves as prompt engineers — meaning that they endeavor to best leverage the data at hand to produce novel, beautiful, interesting or meaningful outcomes. While prompting maintains a sort of poetics, it is also easy to see common vernaculars, tropes and cliches emerging from A.I. tools. The discovery (or simulation) of A.I. creativity has yet to be confirmed, and it seems to be hidden behind a tendency to elaborate the obvious.
This may be due to the use of very general source data, with the desireable characteristic being its breadth. A map is maintained that traces the land that it maps, and from this we get expected elements. Accidents, of course, emerge anyway, a form of chance surealism which requires a human agent working to curate or annotate what has arrived. Even if this allows a poetic rupture to occur in the eyes of the viewer, it is positioned against the mundane, and overtime will disipate. Early AI art is likely to be more wild than that which will come later.
We already notice a kind of inescapable retro-capture in many artworks produced by current tools - images that come from this space strongly bear traces of the past and what has come before it. Mark Fischer called this “hauntology,” and he lamented late capital production’s elevation of the past at the expense of an open future. This is of course is an interesting mirror, but it is always a mirror, and distortions in its view require a kind of critical leap.
An Alternate Approach
An unexplored and perhaps potentiating way to interact with A.I. could be for the artist to curate the dataset, instead. This act could be considered as a world building effort that seeks novelty through interiority, splintering time and protecting anachronism. In a way , this is already a promise of blockchain and the metaverse, which seeks to elevate a form of interior meaning in each DAO, each instance of metaverse structure and block, and these experiments could be enriched with A.I.’s enhancements.
Identifying or creating an artesanal dataset obviates the generalized historicity of mass-market tools. And, it does not limit its potential output. A source of data could be reinterpreted over and over by A.I. and artists to different outcomes, perhaps further enriching the dataset with new findings or adjustment, and maintaining alterations. On a small scale, this intervention would be a true collaboration between A.I. and the artist, occurring within a hermetic arena rather than a collective one, much like most artistic collaboration tends to be (i.e. on a singular canvas, in the context of a film or within a concert).
What’s more, the multivalent quality of this activity (naturally agent-less, or intra-agent) opens it up to the possibility of community interactions to scale the process, catalyzing growth and variety. Consider the world-building exercise of a fan community (i.e. fan-fic), who are likewise working with a finite dataset, and sourcing it for growth and extended creativity. Through a crowd-sourced dynamic, we find more possibility of a dataset’s enrichment and artistic fluidity, allowing for non-narrative, non-finite and ultimately boundless interpretation by a group of people. By training the A.I. iteratively, it would be able to draw any human input back into the central monads established by its cosmology of primitives.
LARPS
LARPs is a fictive project, meaning one which has no forward movement, which hopes to engage with this idea. Its goal is to establish a set of primitives for an unfolding, multilateral world, building a dataset to form. The intricacies in its loose, poetic statements and character attributes are meant to become seeds for future conversations, scenes, epics and soliloquies, interwoven with concrete worldbuilding and archetypal elements. It is bounded by its nascent beings, occupations, costumes and settings, from which might emerge a narrative space of unfolding meaning. What is wonderful about the advent of A.I. text generation (i.e. GPT-3,4, etc.) is the nearly effortless creative energy that might be strung between these monads of curated insight, to form new webs of contextualized derivation.
The role of NFT holders is yet to be established for LARPS, should it emerge as more than an idea, but it can be imagined that, within a system such as I propose, human agents might articulate the connections between ideas, the pace at which new, stanzas, paragraphs and chapters advance old ones, and the collections of elements interacting within any particular story arc. With time, new ideas might be found by the interplay of A.I. and human conversation, and these may become new primitives to be shared.
LARPS on OpenSea
(LARPS NFTs, for now, are meant to be slowly created piece by piece)