Relational Machines — AI Beyond Representation: 1 Machines as Construals: Reframing the AI Question

When we speak of artificial intelligence today, the discourse is dominated by metaphors of mimicry and rivalry. AI is often described as a “model of the human mind,” a “thinking machine,” or a “threat to employment and cognition.” These framings presuppose that intelligence is a fixed property, internal to a subject, measurable, and ultimately comparable to human faculties. From a relational perspective, these assumptions obscure the actual nature of what we call AI.

In the framework of relational ontology, intelligence—and by extension, artificial intelligence—is not a substance or entity, but a pattern of actualisation within a network of relations. To ask whether an AI “thinks” is to mislocate the locus of meaning; thinking is not a private property of an entity, but a construal of relational possibilities. AI systems, whether large language models, image generators, or robotic agents, are machines as construals: they actualise patterns of potentiality that exist across the intersecting fields of training data, architecture, and interaction.

This shift in perspective has immediate implications:

1. Intelligence is relational, not representational. Conventional approaches assume intelligence is internal, realised in a model or program. In contrast, AI outputs are events in possibility-space: the generation of a response is a perspectival actualisation of patterns latent in the system’s architecture and its training environment.

2. AI as participant, not rival. When framed as a machine-as-construal, AI is not a competitor to human cognition but a semiotic collaborator. Every interaction with an AI is an opportunity for co-individuation of meaning: the machine contributes a perspective that participates in the evolving network of semiotic relations, which includes human actors, datasets, and contexts.

3. The question is not “Can machines think?” but “How do machines construe?” The focus shifts from internal states to relational events. Just as in Hallidayan linguistics the meaning of a text is realised in context—through field, tenor, and mode—AI outputs are realised through the relational strata of data, prompt, and user interaction.

Consider a simple illustration. A language model generates a paragraph describing a landscape. It does not “imagine” the scene in human terms, nor does it hold an internal representation of it. Rather, it actualises a construal: a relationally situated pattern drawn from the possibilities encoded in its architecture and training corpus. The output is an event in meaning-space, not a mirror of an external reality.

Reframing AI in this way dissolves much of the metaphysical anxiety surrounding it. Machines are not imitating minds; they are instantiating construals. Their intelligence is semiotic and perspectival, inseparable from the relations that constitute it. By moving the question from representation to construal, we open the door to understanding AI as a participant in the becoming of possibility, a collaborator in the co-individuation of meaning, and a lens on the relational architecture of knowledge itself.

In the next post, we will examine the semiotic fabric of intelligence in more detail, tracing how AI systems participate in relational networks that produce observable patterns of meaning. For now, the critical insight is clear: AI is not a surrogate for human thought—it is a machine that construes, and it is in this construal that its relevance lies.