What if the brain is a computer?
Dive with me into some speculative insights about the brain’s architecture!
Fig. The brain, conceptualized as a von Neumann-style computer, albeit with unique embodied intricacies. (Refresh the page to restart the animation.)
In its high-level organization, the brain bears striking similarities to a von Neumann computer. Here’s a rough analogy:
- Cerebral Cortex: Serves as the primary memory.
- Basal Ganglia: Acts as the control unit or instruction decoder.
- Hippocampus: Can be likened to cache memory.
- Thalamus: Functions similarly to an ALU, albeit focusing on directing data flow. Notably, arithmetic operations seem to occur close to memory, mirroring the design of emerging near-memory computer (NMC) architectures.
- Brain stem: Serves low-level support functions such as the BIOS/southbridge or platform controller hub.
In computer jargon, a computer architect might describe the human brain as an extreme-VLIW (Very Long Instruction Word) symmetric NMC multiprocessor, equipped with a nightly write-back cache and a significant amount of instruction-level parallelism.
If we lean into the metaphor of the brain as a computer, it’s clear that there is complex software operating on this “hardware.” Merely knowing the entire brain’s structural map (often referred to as the connectome) is far from enough to grasp the brain in its entirety. There’s no magic machine learning formula that will unveil its mysteries instantaneously.
What is pivotal to note is that we don’t emerge into the world with this software pre-installed. Instead, it is developed throughout our early years. Unraveling how the brain bootstraps itself—essentially, how it learns to learn or “meta-learning”—might be our best shot at comprehending the higher echelons of the brain’s functions.
For those intrigued by these musings, I invite you to watch my seminar at RISE titled Beyond Deep Learning - What can biology teach us?.
Additionally, a great book discussing the brain by layers of abstraction is Ballard: Brain computation as hierarchical abstraction (MIT Press, 2015).