Foreword 

Why the Transformer architecture fulfills the vision of the Blue Brain Project

When Henry Markram initiated the Blue Brain Project in the early 2000s, many skeptics doubted it. The idea of fully simulating the brain as a numerical model seemed bold, overly ambitious, almost reckless. However, Markram had an intuition that was ahead of its time:

The brain is a rule-based, computable machine. Its architecture is algorithmic. And it can be simulated.

What was missing back then was not the will, the data volume, or the computational power. It was the correct abstraction.

Neuroscience was trapped in biophysical details: ion channels, dendrite models, membrane conductances. Artificial intelligence had not yet developed functional architectures beyond simple feedforward networks. And neuroanatomy was too fragmented to recognize a global functional architecture.

Today, two decades later, the situation has fundamentally changed. With the introduction of the Transformer architecture in artificial intelligence, a model was created that:

• is hierarchical
• is recursive
• is context-sensitive
• operates in parallel
• and exhibits emergent properties

— just like the brain.

Transformers possess:

• Tokens
• Query signals
• Key structures
• Value contents
• Positional encoding
• Multi-level attention
• Recursive loops

And these elements are reflected—anatomically, functionally, and topologically—in the human brain.

Although the highest level of neural signal processing in the human brain is realized through transformer-like architectures, the brain is not purely a transformer.

All fundamental network types known from AI—KNNs, CNNs, RNNs—also exist within the vertebrate nervous system. They form the evolutionarily older layers of signal processing. Their outputs feed into the transformer modules, just as in modern AI systems, preprocessing, feature extraction, and recurrent loops provide input for attention mechanisms.

Nature did not invent these architectures to keep them separate but to combine them. The transformer structures of the human brain are the result of a long evolutionary chain of signal processing systems—and they only work because the underlying networks prepare the signals.

In artificial intelligence, it is no different: transformers only reach their full potential when input signals are structured by other network architectures. AI experts understand this. Biology has practiced it for millions of years.

This makes it clear: Markram's vision was not wrong—it was simply ahead of its time. He sought a numerical architecture capable of functionally modeling the brain. That architecture exists today. It is called Transformer.

The presented theory of biological transformers shows that the brain is not only a biological network but also an algorithmic machine, whose structure is reflected in modern AI models. The numerical simulation Markram aimed for is now possible—not by fully replicating every ion channel but through the functional reconstruction of the brain's signal architecture.

The irony of history is remarkable: while many considered Markram overly ambitious at the time, AI has now produced exactly the models that fulfill his vision. This biological transformer theory demonstrates that nature has been using an architecture for millions of years that we are only now beginning to understand.

Markram wanted to simulate the brain. Today, we can say:

The brain simulates itself—as a transformer.

If the brain is a transformer, then intelligence must be understood as signal processing—not as abstract information processing.

This foreword aims to set the historical context for the subsequent theory. It is not only a neurobiological hypothesis but also an answer to a question that has persisted for decades:

How does intelligence work?

The answer is: as a signal theory—and as a transformer.