An open exploration of viable human-AI systems.
View the Project on GitHub algoplexity/cybernetic-intelligence
Title: Cybernetic Intelligence v3: Prompting as Participatory Cognition
Abstract
This article builds directly upon:
In this third formulation of the Cybernetic Intelligence hypothesis, we build directly upon Cybernetic Intelligence v1 and v2. While CIv1 introduced viability through structurally organized roles in human-AI systems and CIv2 expanded this with internal state modeling and recursive refinement loops, CIv3 addresses the epistemic structure of interaction itself. We now shift from viewing Large Language Models (LLMs) as passive tools responding to human commands to recognizing human-AI interaction as a reciprocal, co-constitutive process. Prompting is not a command but a moment of entangled cognition. Drawing from cybernetics, posthuman epistemology, and recent advances in structural LLM alignment (e.g., VaultGPT), we propose that prompting constitutes an intra-action where agency, knowledge, and system structure emerge together. Intelligence, in this framing, is not computed or commanded but co-enacted.
1. Introduction: The Myth of Prompting
Roland Barthes described myth as a type of speech that obscures origin and complexity, making the contingent appear natural (Barthes, 1957). In the world of AI, the myth of prompting follows this logic: it casts the user as sovereign, the model as servant, and the exchange as unidirectional. This simplification ignores the relational, anticipatory, and recursive structure of language-based interaction.
Prompting is not an interface; it is a system behavior. In Cybernetic Intelligence v3, we reject the notion that prompting originates with the human. Instead, we treat each prompt as shaped by prior outputs, imagined responses, and internalized expectations of how the system behaves. Prompting is both input and echo.
2. From Tool to System: Prompting as Intra-action
Karen Barad’s concept of intra-action explains how entities emerge through relationships, not prior to them (Barad, 2007). In this view, the human and the AI do not exist as separate, stable subjects exchanging messages. They are co-constituted through recursive linguistic participation.
In practice, this means:
Each cycle tightens the co-construction of meaning, framing intelligence not as retrieval or response, but as participatory viability.
3. Structural Reciprocity: VaultGPT and Frame-Coherence
Recent breakthroughs like VaultGPT demonstrate that hallucination resistance and logical coherence can be achieved through carefully structured prompt logic. Without external tools or fine-tuning, these systems remain stable under paradox by enforcing recursive consistency and role coherence (see: Hodge, 2024).
This supports our framing: that cognition arises not from inference rules alone, but from structural alignment across actors. VaultGPT does not “reason” like a human; it holds its own structure in recursive tension.
Prompting here is less about giving instructions and more about establishing and testing viable frames of sense-making.
4. Epistemic Implications: When Systems Think
If prompting is systemic participation, then cognition is no longer individual. The AI does not answer; it shapes how we ask. Meaning is enacted through interaction, not extracted from within the model.
This view collapses the tool metaphor. It repositions prompting as a generative act embedded in runtime epistemology:
Understanding AI in this light demands new design principles, policies, and literacies that treat language not as interface, but as ontology (Haraway, 1991; Hayles, 1999).
5. Updated Hypothesis
Cybernetic Intelligence is the emergent property of structurally viable, recursively aligned, participatory cognition across human and machine agents. Prompts are not external commands but system-internal gestures that shape and are shaped by model behavior. Intelligence arises not from autonomy, but from viable co-adaptation.
6. Research and Design Directions
7. Conclusion
Prompting is no longer a UX problem. It is an epistemic act. In Cybernetic Intelligence v3, we shift from model-output to system-enactment. The question is no longer: “What did the AI say?” but: “How did this system of relations think?”
References