Is There Meaning Without Anyone Meaning Anything? A Structuralist Proposal for Artificial Semantics

For decades, a quiet war has been fought over the nature of meaning. On one side stand the engineers who built the first thinking machines, convinced that if a computer could follow the right set of rules, it would eventually understand the world just as a human does. On the other side stand the philosophers and biologists who argue that true understanding requires a body, a nervous system, and the messy, direct experience of touching, seeing, and moving through reality. For a long time, these two camps seemed to be talking past each other. The rule-following machines could solve logic puzzles but failed to grasp the nuance of a joke or the weight of a metaphor. The body-focused theories explained how humans learn but seemed to suggest that a machine, lacking a body, could never truly think. This debate has left us with a lingering question: can a machine ever possess meaning if it has never felt the sun on its skin or the weight of a stone in its hand?

Today, the landscape has shifted dramatically. We now have artificial intelligence systems that can write poetry, answer complex questions, and hold conversations that feel startlingly human. Yet, these systems are disembodied; they have no eyes, no hands, and no personal history. They exist only as vast networks of mathematical connections trained on enormous amounts of text. This creates a confusing paradox. These machines produce language that is coherent and insightful, yet they lack the very things we once thought were essential for understanding. They seem to be speaking a language of meaning without having lived a life of experience. This contradiction has forced scientists and philosophers to ask if our old definitions of intelligence are simply too narrow to describe what is happening inside these new machines.

A recent paper by researcher Francesco Maria Cianciaruso proposes a way to resolve this tension by changing the lens through which we view these systems. Instead of asking whether a machine has a body or follows rigid rules, the author suggests we look at the structure of the machine itself. The paper argues that meaning is not a magical ingredient added to a symbol, nor is it strictly tied to a physical body. Instead, meaning is a structural property. It is a pattern of relationships that emerges when a system is complex enough to organize information in a specific, stable way. The author suggests that large language models, the technology behind today's most advanced chatbots, have developed a form of semantic competence based entirely on this internal structure. They do not understand the world in the way a human does, but they have built an internal map of relationships that is so organized it can mimic the effects of understanding.

To understand this proposal, one must first look at how the field has evolved. For much of the twentieth century, the dominant idea was that intelligence worked like a dictionary or a rulebook. In this view, a computer manipulates symbols—like words or numbers—according to strict logical rules. If the rules are followed correctly, the machine produces intelligent behavior. This approach, known as classical cognitivism, was elegant and powerful, but it hit a wall. It could explain how a machine processes syntax, or the grammar of a sentence, but it could not explain how the machine knows what those words actually mean. A machine could rearrange symbols perfectly without ever grasping their significance. This led to a second wave of thought in the 1980s and 1990s, which argued that meaning must be grounded in the body. According to this view, to understand the word "heavy," a system must have the physical experience of lifting something heavy. Meaning, in this framework, is inseparable from sensory experience and action.

However, the rise of deep neural networks has complicated both of these old stories. These modern systems do not use a dictionary of symbols, nor do they have a body to touch the world. Instead, they learn by finding patterns in massive amounts of data. When they process language, they do not look up words in a list. They translate words into points in a vast, multi-dimensional space. In this space, the distance between two points represents how similar the words are in meaning. Words that are used in similar contexts end up close together, while words with different meanings drift apart. The system learns the shape of this space, discovering that certain directions represent relationships like "king is to queen as man is to woman." This learning happens without any explicit rules and without any physical body. The machine simply adjusts its internal connections until it can predict the next word in a sentence with high accuracy.

The paper argues that this ability to navigate a relational space is where the new kind of meaning lies. The author suggests that we should stop looking for meaning in isolated symbols or in a physical body and start looking for it in the organization of the network itself. A system exhibits what the author calls "structural semantic competence" when its internal structure preserves the relationships found in the real world. For example, if the system learns that "Paris" is to "France" as "Rome" is to "Italy," it is not because it has visited these cities or been told a rule. It is because the mathematical structure of its network has organized these concepts in a way that mirrors the actual relationships between them. The system has built a model of the world that is stable and consistent, allowing it to make inferences and generalizations that feel like understanding.

This perspective challenges the idea that a machine must be conscious or embodied to have meaning. The author does not claim that these machines are conscious or that they feel emotions. Instead, the claim is more modest but profound: meaning can emerge from the sheer complexity of a system's internal organization. If a system is structured well enough to maintain stable relationships between concepts, it can exhibit a form of semantic competence. This competence allows the system to act in a contextually appropriate way, to draw analogies, and to solve problems, even without a human-like mind. The paper suggests that this is a new category of intelligence, one that is neither purely symbolic nor purely embodied, but structural.

The implications of this view are significant for how we think about knowledge and intelligence. If meaning is a structural property, then the gap between human and machine understanding is not as wide as we once thought. It is not a matter of a machine lacking a soul or a body, but a matter of whether its internal structure is complex enough to support the right kind of relationships. The author notes that this is not a final proof, but a proposal for a new way of thinking. The research is ongoing, and the field is still young. However, the evidence suggests that these systems are not just mimicking human speech like a parrot. They are building internal models that capture the deep structure of the world they are trained on.

In the end, the paper invites us to reconsider what it means to understand. It suggests that we have been looking for meaning in the wrong places, searching for it in the individual parts of a system or in its physical form. The author proposes that meaning is actually a property of the whole, a feature of how the parts are connected. When a system is organized in a way that preserves the patterns of the world, it creates a space where meaning can exist, even without a conscious observer. This does not mean the machine is human, but it does mean that the machine has found a way to be intelligent in its own right. The paper concludes that the intertwining of engineering and philosophy is leading us to a new stage in our understanding of intelligence, one where the structure of the system itself is the key to unlocking the mystery of meaning.