For almost two hundred years, neuroscience has narrated a narrative about the human brain that revolves predominantly around a singular type of cell.
The neuron — the branched, electrically-active cell responsible for transmitting signals across synapses — has been the highlight of every popular portrayal of thought, education, and remembrance since the mid-1800s. The well-known figure of 86 billion neurons in the human brain appears in every textbook, every scientific documentary, every informal discussion about what contributes to our intelligence. This number implies a hidden claim. The reason humans possess the ability to think, learn, and recall at our level is due to having this specific vast quantity of these particular electrical cells.
What is less frequently brought to light is that the brain also harbors a roughly similar number of a wholly different type of cell — possibly around 85 billion of them, depending on the counting method — which neuroscience has largely regarded as more or less inconsequential.
These are called astrocytes. Their star-shaped appearance gives them their name. They feature long, branching extensions that reach out over extensive distances within the brain, interacting with millions of neurons each. And for the majority of its existence as a discipline, modern neuroscience has overlooked them as mere biological packaging — the wet, structural filler between the more fascinating cells.
A paper published in the Proceedings of the National Academy of Sciences in May 2025 suggests that this oversight may represent one of the most significant mistakes in neuroscience history.
The origins and disappearance of astrocytes
The term “astrocyte” was introduced in 1893 by the Hungarian anatomist Mihály Lenhossék. He built upon several decades of prior observations made by researchers such as Rudolf Virchow, the innovative German pathologist who, in the 1850s, designated the family of non-neuronal brain cells as glia — derived from the Greek term for glue.
This nomenclature encapsulates much of how the field has perceived these cells for the ensuing century and a half. Neurons constituted the brain’s essence. Glia — including astrocytes — served as the adhesive providing cohesion to that essence. They were intriguing, akin to scaffolding being interesting. Not the elements one aimed to comprehend.
This assumption solidified throughout the twentieth century as neuroscience increasingly concentrated on electrical activity as the currency of brain function. Neurons fire. Neurons signal. Neurons establish and reestablish the patterns that form learning and memory. Astrocytes do not exhibit this behavior in any evident manner — they do not generate electrical action potentials, they do not convey signals across synapses as neurons do, and they typically do not engage in the rapid electrical dynamics that facilitate neural computation.
The logical deduction, reached by successive generations of neuroscientists, was that astrocytes functioned as background infrastructure. Blood-brain barrier preservation. Ion equilibrium. Nutrient provision to eager neurons. Support roles.
Now, it appears this reasonable conclusion was incorrect.
The MIT team’s current proposal
The 2025 paper was spearheaded by Leo Kozachkov, collaborating with Jean-Jacques Slotine of MIT and Dmitry Krotov of the MIT-IBM Watson AI Lab. Their proposition is technical, yet the core concept is straightforward.
Astrocytes, they contend, are not merely passive observers. They are computing entities.
Each astrocyte reaches out its long branches to connect with an immense number of neurons — often millions at a time. At each of these connections, the astrocyte is capable of chemically influencing the activity of the neuron, releasing neurotransmitters, balancing ion levels, and modulating the strength of individual synapses. This process is not the rapid electrical signaling observed in neurons themselves. It is more gradual, subdued, and distributed — a form of parallel computational layer existing alongside the neural network that has been under scrutiny.
If the MIT team’s assertions hold true, then the brain comprises not one, but two intertwined information-processing systems. The neuronal framework, functioning with electrical signals across billions of connections, and the astrocytic framework, operating with chemical modulation across billions of contact points. They are not distinct entities. They intertwine — every neuron in your brain is currently influenced by astrocytes that are determining its behavior in the upcoming moments.
The MIT group applies a model of this duality using a framework known as dense associative memory — a mathematical construct that can, in theory, retain significantly more information than a model predicated solely on neurons. The specific design of the model aligns with what is known about astrocyte connectivity in a way that seems improbable by mere coincidence.
Understanding the nature of memory scaling
The most immediately notable implication of the model is that it may elucidate a mystery that has perplexed neuroscientists for years — the human brain’s seemingly limitless capacity for memory.
A conventional neural network, the model that has guided our understanding of memory since the 1950s, possesses a distinct storage capacity. The number of unique patterns it can maintain before the patterns begin