Traditional AI encodes memories in the synaptic weights between neurons. However brains are more than just neurons. The number of brain glial cells is approximately equivalent to the number of neurons.
Recent research has focused on astrocytes, the most abundant type of brain glial cell. Astrocytes tile the brain and wrap their processes around individual synapses. They are electrically excitable and contain potassium and calcium channels in addition to gap junctions. One astrocyte can connect itself with over 100,000 synapses.
Now, researchers at MIT have proposed a new model of associative memory in neural networks, leveraging astrocytes. This model increases memory storage by a factor of N, the number of neurons. In brains or in an AI, this number is large, it could be up to 100 billion.
Recent research has focused on astrocytes, the most abundant type of brain glial cell. Astrocytes tile the brain and wrap their processes around individual synapses. They are electrically excitable and contain potassium and calcium channels in addition to gap junctions. One astrocyte can connect itself with over 100,000 synapses.
Now, researchers at MIT have proposed a new model of associative memory in neural networks, leveraging astrocytes. This model increases memory storage by a factor of N, the number of neurons. In brains or in an AI, this number is large, it could be up to 100 billion.
