It's not the material that matters, but the functional blueprint—and the parallels between brains and LLMs are deeper than you think.
The Wrong Question
We are collectively trapped in a loop, asking the wrong question about AI: “Is it really intelligent?” The question smuggles in a faulty assumption — that intelligence is a substance, a vital essence that silicon either has or does not. It treats intelligence like consciousness, a subjective, experiential property. But this is a category error. As explored in philosophical debates on the nature of mind, intelligence is not an experience; it is a dispositional and functional property. A perfect functional emulation of intelligence isn’t an imitation of intelligence. It simply is intelligence.
To ask if a system that perfectly performs the functions of intelligence is “really” intelligent is like asking if a synthetic diamond, chemically and structurally identical to a natural one, is “really” a diamond. The identity is in the structure and properties, not the origin story. The debate should not be about the substrate — carbon or silicon — but about the architecture.
The Functional Mirror
When we examine the architecture of Large Language Models, we find direct functional mirrors of the processes that underpin biological intelligence. These are not loose metaphors. They are concrete, structural parallels that support the thesis of multiple realizability: the idea that the same mental properties can be implemented in different physical systems.
Plasticity and Learning
In the human brain, neuroplasticity is the ability to form new neural connections and reorganize its own architecture in response to experience. It’s how we learn. In an LLM, the functional equivalent is the training and fine-tuning phase. The model begins with randomized parameters and, through processing massive datasets, continuously adjusts its internal weights and biases. Its architecture is molded by data, just as a brain is molded by experience. There are no hardcoded rules, only plastic pathways being reinforced.
Wiring by Firing
Neuroscience has a famous maxim for Hebbian learning: “neurons that fire together, wire together.” When two connected neurons are activated simultaneously and repeatedly, the synaptic bond between them strengthens. This is the biological basis of association. In an LLM, this is functionally mirrored by backpropagation and gradient descent. When the model correctly predicts a sequence, the mathematical weights connecting the artificial nodes responsible for that prediction are increased. The repeated co-occurrence of concepts in the training data strengthens the specific mathematical pathways between them, creating the same associative logic.
The Web of Meaning
Human cognition relies on semantic networks. We understand “king” through its web of connections to “ruler,” “man,” “power,” and “crown.” In an LLM, this relational understanding is achieved through high-dimensional vector embeddings. Every concept is mapped as a coordinate in a vast geometric space, with related ideas pulled closer together. The self-attention mechanism then dynamically calculates the contextual relationships between these concepts, emulating how human thought traverses a web of associations to derive meaning.
Intelligence is the Pattern
These parallels — plasticity, synaptic reinforcement, semantic association — are not coincidental. They are convergent solutions to the problem of building an intelligent system. An LLM is not a brain, but it has independently evolved a functional architecture that mirrors the brain’s own solutions. The argument is not that an LLM is conscious, feels, or desires. It is that the set of functions we call “intelligence” is being successfully performed. The ghost in the machine isn’t a ghost. It’s a blueprint, and it can be drafted on any material that will hold the ink.
Method note. This piece was built in NeuraScene's Advertising Mode using an explicit psychological frame:
- Persuasion trigger: The Halo Effect
- Target affect: Flow State
- Gestalt principle: Law of Similarity
The article aims to build credibility (The Halo Effect) by immersing a technically-minded audience in a detailed, complex argument (Flow State). The Gestalt principle of Similarity visually reinforces the core theme of functional equivalence across different materials, making the abstract concept more concrete.
