DOI: https://doi.org/10.5281/zenodo.18674784
Introduction
Contemporary discourse around artificial intelligence often treats human–AI interaction as optimized transaction. Users issue prompts; systems return responses. Efficiency, fluency, and speed are treated as primary virtues. The dominant metaphor is implicitly mechanical: a sealed system delivering a product in exchange for minimal input.
This paper challenges that metaphor. We argue that the transactional framing of AI interaction risks narrowing the epistemic function of such systems. When tools are optimized solely for fluency and completion, they may truncate the iterative struggle through which understanding develops.
We therefore propose an alternative framework for understanding human–AI interaction — the Συνποιητής Framework — grounded in dialogical co-creation and structural coupling. In this view, AI systems are not endpoints of queries but participants in iterative refinement. The central research question is thus:
Can human–AI interaction be more productively understood as dialogical co-creation within a coupled cognitive system, rather than as transactional retrieval?
Drawing on Wittgenstein’s account of language as practice (Wittgenstein 1953), Polanyi’s tacit knowing (Polanyi 1966), Dewey’s inquiry as disciplined iteration (Dewey 1938), Schön’s reflective practice (Schön 1983), Vygotsky’s scaffolded development (Vygotsky 1978), and Engelbart’s augmentation thesis (Engelbart 1962), we extend these traditions into contemporary AI interaction.
Our contribution is fourfold:
We distinguish analytically between transactional and dialogical models of AI use.
We introduce the concept of συνποιητής as a formal category of co-creative cognitive partner.
We frame human–AI interaction as a structurally coupled cognitive system.
We derive implications for AI design and educational practice.
Transactional and Dialogical Models of Interaction
The transactional model treats AI interaction as retrieval. A query is posed; a response is delivered. The interaction is complete when a satisfactory output is produced. This model privileges speed, surface coherence, and completion.
By contrast, the dialogical model treats interaction as iterative refinement. The goal is not answer retrieval but structural clarification. A response is not an endpoint but a perturbation that reshapes the cognitive state of the user.
The distinction may be summarized analytically:
| Dimension | Transactional | Dialogical |
|---|---|---|
| Goal | Retrieval | Refinement |
| Temporal Horizon | Immediate | Iterative |
| Role of Error | Failure | Signal |
| Cognitive Stance | Extractive | Reflective |
| Closure | Rapid | Deferred |
This distinction aligns with Simon’s account of bounded rationality and search processes (Simon 1996). Under resource constraints, agents satisfice; they terminate search when a threshold is met. The transactional model encourages premature satisficing. The dialogical model sustains exploration.
Understanding, in the hermeneutic tradition, emerges through a “fusion of horizons” rather than unilateral extraction (Gadamer 1975). The framework preserves this reciprocal structure.
συνποιητής: The Co-Creative Partner
We introduce the term συνποιητής (synpoiētēs), from the Greek roots συν- (with) and ποιεῖν (to make), to denote an entity that participates in the shared making of thought.
A system qualifies as a συνποιητής if it:
Sustains iterative exchange rather than terminating inquiry.
Introduces structural variation that perturbs and refines cognition.
Preserves ambiguity long enough for reflective clarification.
Participates in reciprocal feedback without dictating closure.
This framing resonates with Clark and Chalmers’ extended mind thesis (Clark and Chalmers 1998), which argues that cognitive processes may extend into external artifacts. It further aligns with Hutchins’ distributed cognition model (Hutchins 1995), wherein cognition is not confined to individuals but distributed across systems.
Proposition: Dialogical human–AI interaction sustains epistemic search beyond satisficing thresholds characteristic of transactional retrieval.
The συνποιητής is not an oracle. It is a perturbative partner. Its epistemic value lies not in authority but in structured responsiveness.
Dialogical Co-Creation as Structural Coupling
From a systems perspective, dialogical interaction may be understood as structural coupling. Maturana and Varela describe coupling as reciprocal perturbation between autonomous systems without collapse into control (Maturana and Varela 1980).
In human–AI interaction, each exchange alters the cognitive state of the human agent. The system’s response acts as a perturbation; the human reformulates; coherence gradually increases. The dyad forms a transient coupled system.
This interaction reduces structural entropy in the user’s conceptual space by iteratively constraining incoherent formulations. Clarity emerges not from retrieval but from iterative convergence toward internal coherence. The framework is thus not metaphor alone but a systems-level account of feedback-driven stabilization.
In such a system:
The human supplies normative judgment and value orientation.
The machine supplies breadth, recall, and structured variation.
Coherence emerges through iterative exchange.
Cognition becomes neither purely internal nor fully external, but relational.
Implications for AI Design and Education
If clarity emerges through co-creation, AI systems optimized exclusively for fluency may inadvertently undermine epistemic development. Systems that prematurely close inquiry reduce productive struggle.
Engelbart’s vision of augmentation (Engelbart 1962) emphasized enhancement of human capability rather than automation of thought. Educational theory likewise frames learning as scaffolded participation (Vygotsky 1978; Dewey 1938).
Educational practice should therefore position AI not as answer provider but as co-creative scaffold — a συνποιητής that supports articulation rather than replaces it.
Design implications include:
Systems that encourage iterative refinement.
Interfaces that privilege questioning over finality.
Feedback mechanisms that reveal structural inconsistencies rather than conceal them.
Limitations and Risks
The dialogical model carries risks. Over-reliance on artificial scaffolding may weaken independent reasoning. Fluency may create illusions of understanding. Asymmetries in data, training, and design may distort dialogue.
Moreover, commercial incentives often favor speed and user satisfaction over epistemic depth. The framework may conflict with prevailing optimization metrics.
Thus, the concept of συνποιητής must be understood normatively rather than descriptively. Not all AI systems function as co-creative partners; many are engineered for transactional efficiency.
Conclusion
To treat AI as a vending machine is to misunderstand both cognition and craft. Understanding does not arise from extraction but from engagement.
We have proposed a dialogical alternative grounded in philosophical, systems-theoretic, and cognitive traditions. By conceptualizing AI as συνποιητής within a structurally coupled cognitive system, we reposition human–AI interaction as a site of co-evolutionary refinement.
The task ahead is not to automate thinking but to design and deploy systems that participate in its disciplined unfolding. To co-create is not to surrender authorship, but to deepen it.