A study published in the journal Science has upended 80 years of conventional wisdom in computational neuroscience that has modeled the neuron as a simple point-like node in a system, integrating signals and passing them along. This neuron-as-integrator model, also known as the “dumb” neuron model, has severely restricted the conception of what a neuron is capable of doing, and hence how neuronal networks and the brain as a whole functions.
This has not only impeded the development of a complete understanding of neuronal activity in the higher brain regions of the cortex, but it has also adversely affected computer science, significantly limiting the development of neuromorphic computational networks because they have been based on an incomplete model. Empirical investigations are now suggesting that scientists re-evaluate neuronal information processing as a much more complex system—one that may not have direct parallels with our computational technologies.
The Complexity of Neuronal Information Processing
The new research by Professor Matthew Larkum, a neuroscientist at Humboldt University, and his team has uncovered a rarely before seen information processing system in single dendrites of pyramidal neocortical neurons that uses graded signal processing with calcium-mediated dendritic action potentials, as opposed to the typical all-or-nothing action potentials observed in sodium and potassium ion flux. (Electrical current is conducted through the nervous system, referred to as an action potential, via flux of large cations like sodium, potassium, magnesium, and calcium.)
Dendrites are branched protoplasmic extensions of the cellular membrane of the neuron. Like a tree forming branches from the trunk, dendrites are arborizatons from the neuron soma that contain the synaptic architecture necessary to receive, process, and transmit electrical signals (from the axons of adjacent neurons). In some classes of neurons, there are thousands of dendrites, and when summing all of the subsynaptic structures, a single neuron can form as many as 100,000 signal processing / integrating connections—so extensive that when modeling such connectivity one research team utilized a mathematical manifold of 11-dimensions.
What the recent research has discovered is that the dendrite is much more than a simple receptor and integrator of signals. There is a complex sub-synaptic architecture that gives the dendrite the processing power normally attributed to a multi-layered neuronal network—i.e. the dendrite alone can perform complex computations, and hence the multi-parallel processing power of a single neuron is far beyond what was conventionally assumed.
When commenting on the model of the neuron as a simple integrator, Bartlett Mel, a computational neuroscientist at the University of Southern California, said: “That’s essentially the neuron being collapsed into a point in space. It didn’t have any internal articulation of activity. The model ignored the fact that the thousands of inputs flowing into a given neuron landed in different locations along its various dendrites. It ignored the idea (eventually confirmed) that individual dendrites might function differently from one another. And it ignored the possibility that computations might be performed by other internal structures.”
The Implications for Computational Neuroscience and Technology
The new discovery confirms a prediction made by the science research group at Torus Tech LLC. In my own model discussing the molecular cytoarchitectonics of the brain and its role in consciousness, I had described the new neurocomputational paradigm as follows (note that my description of information processing in the biological system is pointedly not limited to neurons):
“The scale-free complexity associated with the biological system in general, and the neuron in particular, means that within each cell there is a veritable macromolecular brain, at least in terms of structural complexity, and perhaps to a certain degree functional complexity as well—a fractal hierarchy. This means that the extremely simplistic view of the synapse as a single digital bit is misrepresenting the reality of the situation—such as, if we were to utilize the parlance of the neurocomputational model, each ‘computational unit’ contains a veritable macromolecular brain within it. There is no computer or human technology yet equivalent to this.”
– William Brown, Resonance Academy Big Questions Course, Lesson III: The Cellular Hologramic Information Nexus | Sentience and Memory Encoding in Cellular and Macromolecular Systems. 2018.
The model being described in my course The Cellular Hologramic Information Nexus is essentially saying that the computing power normally attributed to the brain as a whole is probably contained within a single neuron. And indeed, the latest research confirms that there is multi-layered information processing occurring in single neurons, and that is only evaluating the sub-synaptic structures of the dendrite. When extended to the internal structures of actin filamentary networks, Posner clusters, and the mitochondrial reticular matrix, which may be operating with quantum principles for massive parallel processing, the computational capacity of a single cell will be found to be staggering.
Gordon Shepherd at the Yale School of Medicine has stated as much when he said, “Much of the power of the processing that takes place in the cortex is actually subthreshold; a single-neuron system can be more than just one integrative system. It can be two layers, or even more.” In theory, almost any imaginable computation might be performed by one neuron with enough dendrites, each capable of performing its own nonlinear operation.
Beyond the “Cartoon Neuron”
Stuart Hameroff’s pioneering work, as presented in his 2023 symposium “Neuroscience Needs a Revolution,” directly challenges what he identifies as the fundamental problem in current neuroscience: the oversimplified “cartoon neuron” model. As Hameroff argues, “The main problem in not understanding consciousness is… in our neuroscience assumptions.” Conventional approaches treat neurons as simple computational units where “each neuron or its firing is considered a bit or a fundamental unit.” This model, Hameroff contends, is “an insult to neurons,” as each neuron is “incredibly complex” with “12 orders of magnitude of dynamic activity.” Hameroff’s collaborative work with Roger Penrose on Orchestrated Objective Reduction (Orch OR) theory proposes that microtubules within neurons—rather than just synaptic connections—are key to information processing, particularly in pyramidal cells which he identifies as “the most likely site for consciousness in the brain.” His calculations suggest that microtubule-based processing yields “10^16 operations per second per neuron”—matching what conventional approaches attribute to the entire brain. This quantum perspective extends processing from mere hertz frequencies to “terahertz to gigahertz to megahertz,” creating a scale-invariant hierarchy that moves “quantum-friendly” processing deep within neuronal architecture. Such findings demand we reconceptualize the brain “not as a computer, but something like a quantum orchestra,” with profound implications for our understanding of consciousness, memory, and neurological disorders—and potentially pushing “the goalposts for AI way downstream.”
Reference:
Dendritic action potentials and computation in human layer 2/3 cortical neurons.
BY ALBERT GIDON, TIMOTHY ADAM ZOLNIK, PAWEL FIDZINSKI, FELIX BOLDUAN, ATHANASIA PAPOUTSI, PANAYIOTA POIRAZI, MARTIN HOLTKAMP, IMRE VIDA, MATTHEW EVAN LARKUM
SCIENCE03 JAN 2020 : 83-87
Dendritic action potentials extend the repertoire of computations available to human neurons.
