It’s been said that human brain has a multi-level cache similar to a modern computer. While a modern computer has multiple levels of cache, CPU Lx cache, memory, disk, tape… The multi-level cache designs offer a convenient way to access data; physically close caches are faster to access, but are more likely to have a limited capacity, due to shear physical reasons. The human short memory is extremely fast, and works like a limited LRU queue, where new sensory data replaces relatively older data. It’s been said that the magic retention index is 7 +/- 2. If there are more than 10 data points in the short memory, the older points are “dumped” to the long term memory; aka “Human memory”. Compared to a network of interconnections -rather than data banks- the long term memory is inaccurate to access, but often survives for many years. It has been proven that different data types are stored in different parts in the brain, (speech, images, shapes).
Compared to a CPU, the neural interconnections are sluggish: 100 meters per second, and a maximum of 1000 clock cycles per second, but the parallel design of the neurons is staggering: 10 billion neurons x 25,000 connections per neuron. The parallel nature of the neurons provides a fault-tolerant system capable of solving parallelizable problems -such pattern recognition- in a blink of an eye. However, tasks that cannot be pararallelized like arithmetics are not easily processed. Consequently, the brain is well designed to process probabilistic, rather than deterministic data; since the data is stored as compressed “icons” as opposed to a lossless image.