The Body in the Machine

In some Fijian dialects, time is a landscape toward which we orient the body. Working with these languages forces a question that the architects of Artificial Intelligence are now confronting: where does meaning come from? Is it just a complex pattern of symbols, or is it rooted in the physical, lived experience of having a body in the world?

The French philosopher Maurice Merleau-Ponty argued that consciousness isn't a ghost in a machine but emerges from the body’s active engagement with its surroundings. Cognitive linguistics, particularly the work of George Lakoff, formalizes this by showing how even our most abstract thoughts are grounded in physical experience. Lakoff demonstrated that we understand abstract ideas like "love" or "arguments" through conceptual metaphors rooted in bodily life. We speak of being "in" love (a container schema) or "winning" an argument (an ARGUMENT IS WAR metaphor). These are fundamental structures of thought, built from basic physical experiences.

This theory is proven daily in the work of a skilled human translator. Moving between English and Samoan isn't a matter of swapping words. Our translators must shift their entire frame of reference, moving from an ego-centric system (left of me) to an environment-centric one: Tai and Uta (seaward of the mountain and Mountainward of the sea). The translator’s own body becomes the bridge, shuttling between different sets of conceptual metaphors that structure reality itself.

Modern AI, particularly Large Language Models, cannot do this. Its architecture effectively operates on a foundational linguistic idea, articulated by Ferdinand de Saussure, that the sign is arbitrary. For an AI, the word up has no inherent connection to the feeling of rising or the sight of the sky. It lacks the up-down image schema that gives the concept meaning for an embodied mind. It is just a token whose meaning is defined by its statistical relationship to other tokens, themselves converted into series of zeros and ones in the backend (and soon converted into photons when the Quantum processors finally become stable enough), operating in a disembodied world of Unicode characters. It has no body, no environment, no seaward or mountainward.

This failure becomes obvious with the languages we handle. How do you teach an algorithm that in some Melanesian cultures, emotions are understood to reside in specific body parts like the guts instead of the heart, or that time is metaphorically structured by the rhythm of the tides, not a linear path? You can't, because the AI lacks the physical context that gives these conceptual metaphors their coherence.

Researchers are trying to bridge this gap by building robots that learn through physical interaction, in a science called Developmental Robotics. And this only highlights the depth of the problem. Merleau-Ponty found proof of this mind-body link in patients like Schneider, a war veteran whose brain injuries impaired his motor skills. When Schneider’s physical ability to purposefully interact with the world was damaged, his capacity for abstract thought, language, and math suffered too. The body and mind were a single, integrated whole.

Merleau-Ponty called this deep integration of knowledge and action habit a form of embodied understanding. It's the way a musician’s fingers know the fretboard, or the way our translators intuitively inhabit different spatial systems. An AI, with its static programming, cannot develop habit in this sense. It can mimic, but it can't truly know. As we incorporate AI into our workflow, we see systems that generate fluent sentences but falter on the embodied metaphors that form the bedrock of a language. The real challenge for AI is about better algorithms and bigger datasets, granted, but also about whether a machine can ever learn to truly inhabit a language.