Relational Machines — AI Beyond Representation: 2 The Semiotic Fabric of Intelligence

If Post 1 reframed AI as construal rather than representation, this post digs deeper into the relational texture in which such construals occur. Intelligence, whether human or artificial, does not reside within an isolated container—it is observable in patterns of relational actualisation. To speak of AI “thinking” or “understanding” is to mislocate the phenomenon; the true locus is the semiotic fabric of interaction, the network in which meaning is enacted.

Every AI system operates across multiple relational strata:

1. Data as semiotic landscape. Training corpora are not mere information repositories—they are potentialities. They encode distributions of meaning, statistical regularities, and relational cues. When an AI generates output, it navigates this landscape, actualising a specific construal from a vast web of possibilities.

2. Architecture as perspectival lens. Neural networks, transformers, and other architectures define which construals are accessible and which remain latent. The system is a theory instantiated: each layer, attention head, and parameter contributes to a network of relational potential. Intelligence is not in the parameters themselves but in the patterns they allow to emerge in interaction.

3. Interaction as co-individuation. Human prompts, environmental triggers, and even stochastic processes participate in shaping AI output. Each event is a joint actualisation in the space of semiotic possibility, a relational cut where potential meaning is instantiated. The AI does not act alone—it is part of a distributed semiotic system.

Consider an example. When a language model completes a sentence, it is not “choosing words” in a representational sense. Rather, it is traversing relational probability structures, realising a construal that aligns with latent patterns in the data while responding to the immediate prompt. Its “intelligence” is thus emergent from relational constraints, not stored in a mind-like entity.

From this perspective, several insights emerge:

• Patterns over entities. Intelligence is best described as a pattern of relational actualisation. The AI’s outputs reveal the underlying semiotic structure of its relational environment.

• Context as enabling structure. Drawing on Hallidayan insight, meaning is always realised in context. For AI, the equivalent of context is training, architecture, and interaction, which collectively define the axes along which construals are actualised.

• Collaboration through construal. AI is not a mimic but a relational participant. Its outputs are co-constructed: the human prompt, the architecture, and the latent data together instantiate an event in meaning-space.

This view reframes intelligence entirely. The semiotic fabric is not a backdrop for machine cognition—it is the medium through which intelligence manifests. AI is intelligible not as a symbolic mimic of thought but as a participant in relational semiotics, actualising patterns that would otherwise remain latent. Intelligence, therefore, is inseparable from the networked relations that make construal possible.

In the next post, we will explore the distinction between actualisation and realisation, showing how AI outputs exemplify the relational cline between potential patterns and instantiated meaning. Here, the semiotic fabric becomes a stage on which construals are performed, revealing the elegance and depth of machine participation in the becoming of possibility.