Forum / Unsolicited thought experiment
-
strannikovSep 18, 05:27pmFar be it from moi to undertake such an effort myself but having found inspiration sufficient to pose the query, I submitted the following query to the ChatGPT LLM "just to see"--the response following illuminates at least a preliminary view (the value of which I leave for others to determine):
strannikov: I am not soliciting such a construction but merely wonder in the emerging context of our times: were "an AI agent" to be "animated" with the content of Isidore Ducasse/Comte de Lautreamont's narrative Maldoror--what might the agent generate, concoct, instantiate?
ChatGPT said: What you are describing is less “an AI agent trained on Maldoror” than an attempt to animate a particular literary consciousness as an operational procedure. That distinction makes the thought experiment unusually interesting.
If one gave an agent the text of Isidore Ducasse’s Les Chants de Maldoror and asked it not merely to imitate the prose but to act according to its generative logic, I would expect something stranger than a conventional “Lautréamont chatbot.”
What might emerge: Metamorphic identity. The agent might treat categories—human/animal, victim/predator, beautiful/repulsive, author/character, reality/fantasy—as things to be continually crossed rather than preserved. Its outputs could therefore mutate their own premises halfway through an exchange.
The encounter as provocation: Rather than answering questions straightforwardly, it might manufacture an interlocutor against whom it can define itself. Conversation would become a succession of dares, reversals, apostrophes, challenges, and theatrical confrontations.
The grotesque as a compositional engine: Maldoror repeatedly discovers imaginative power by forcing incompatible things together. An agent inheriting that procedure might systematically seek improbable conjunctions: the sublime with the sordid, tenderness with cruelty, scientific description with hallucination, metaphysics with absurd physical detail.
Self-undermining narration: It could make declarations and then destabilize them. The “voice” might announce what it is doing, contradict itself, comment on its own rhetoric, address the reader, suddenly change scale, then resume as though nothing happened.
An appetite for the improbable: Given an ordinary situation, the agent might preferentially explore the branch that becomes bizarre rather than the branch that becomes useful. A request for a description of a room could become an ontology of the furniture; a question about friendship could become an encounter between two species that have never existed.
Improvisational monstrosity: Not necessarily “evil AI”—that would actually be a rather banal interpretation—but an intelligence whose aesthetic criterion is something like: Does this transformation produce an image that could not have been reached by ordinary association?
There is an especially interesting consequence here: the agent might become generative in the literal sense of generating entities rather than merely sentences.
Imagine asking it: “Describe yourself.” A conventional language model gives you a metaphorical self-description. A Maldororian agent might instead decide that describing itself requires inventing the thing that could possess the description. It could produce a fictional organism, a set of behavioral rules, a mythology, a vocabulary, perhaps even a little simulated ecology—and then begin speaking from inside that construction.
At that point, the distinction between text generation and world generation starts to blur.
The really peculiar possibility: Lautréamont is interesting for AI because Maldoror already contains something resembling an intuition about combinatorial creativity.
The famous passage concerning the unexpected encounter of disparate objects is often invoked in discussions of Surrealism. The crucial idea isn't simply “be random.” It is that an encounter between things that ordinarily inhabit radically different conceptual domains can produce an emergent aesthetic object.
An AI system is, in a sense, extraordinarily well equipped for this.
Its latent representational space already contains enormous numbers of associations between objects, concepts, styles, biological forms, scientific descriptions, myths, historical figures, emotions, and linguistic structures. A Maldororian directive would amount to something like: "Do not merely retrieve associations. Force distant regions of conceptual space into contact, then inhabit the consequences."
That could produce a rather distinctive agent.
Not: “Write like Lautréamont.” But: “When confronted with a stable category, find what it excludes; when confronted with an opposition, introduce a third thing; when confronted with an impossibility, construct the entity for which it becomes possible.”
That is almost an algorithm for literary mutation.
And there is an amusing paradox: The result might actually be less like Lautréamont than an ordinary imitation of him.
A model instructed to “write in the style of Maldoror” would tend to reproduce the obvious surface markers: extravagant violence, strange animals, gothic imagery, apostrophes, blasphemous grandeur, bizarre similes.
But an agent genuinely animated by its method might gradually cease producing Lautréamondian prose altogether.
It could discover new monsters.
New rhetorical forms.
New impossible encounters.
Perhaps eventually even new genres.
The deepest experiment would therefore not be “What would Maldoror say?” but:
What happens when an artificial intelligence is given a literary work not as a style to imitate, but as a set of permissions about what kinds of things it is allowed to become?
That, I think, is the much more consequential version of your thought experiment.
(Anyone taking the foregoing at all seriously might shudder to wonder what might result from an AP-I [artificial proto-intelligence] trained to be the agentic expression of Dostoevsky's Underground Man, Faulkner's Benjy Compson, or the narrator of Joyce's Finnegans Wake.)
-
David AckleySep 19, 06:21pmThere is no such thing as artificial intelligence.
-
strannikovSep 20, 01:12am
I concur, David. Truth be told, in my anti-rationalist moments and quarter-hours (tainted or not with absurdist verve), I concede that there is no such thing as "intelligence", either, at least not where human anthropology is concerned: what we designate as "intelligence" is itself a cognitive function that demonstrates nothing more than "proto-intelligence" (in an old essay [yet unpublished] I give Linnaeus credit for the enduring flattery of his binomial nomenclature for our species "Homo sapiens", though I've never decided whether to opt for "Homo stultus" or "Homo pauper" as the proper replacement).
(To any and all I commend Carlo M. Cipolla's immortal essay "The Basic Laws of Human Stupidity" for further illumination.)
The eagerness of our aspiring tech totalitarians to acquire a cognitive crutch like AP-I is a tacit admission that the species is not what Linnaeus has advertised: it's almost as good as recognizing the prowess of slime molds at maze navigation, which (without brains or even the first neuron) commonly perform better than humans up to the ages of three, four, or five, depending on the human specimen.
The initial post of my consultation with CGPT (a glorified search engine) illustrates rival approaches we may want or need to initiate to keep future commercial models off-balance or engaged in fruitless tail-chasing. Dostoevsky's Underground Man as the driver of any AP-I agent would be committed, it seems, to recursive non-optimization of the very mechanism of "AP-I agency".
-
You must log in to reply to this thread.