Brain science vs. AI. Why you say Spike NN is like your brain in terms of working mechanism?
Today, I visited a friend. He talked about his ambitious project but what really impressed me is that he insist the similarity in working between brains vs. Spike Neural Networks (idle states of neurons, working in parallel, low energy, etc.)
Uhmm… I read about SNN before but it failed to impress me because of its low popularity. So, today, again I heard about it and this made me curious: Are we already understanding brains?
I sat down and googled, then the first thing came to me is this statement from Christof Koch, Ph.D., Chief Scientist and President of the Allen Institute for Brain Science (this article is published in March 2019):
“We don’t even understand the brain of a worm,”
Hah! Interesting! So we have a long way to go, neuroscientists say.
Reading further, there are 5 big questions in brain science now, but the concerns that seems to be most related to AI (or Neural Networks) are:
- What is the brain made of?
We are building Artificial NNs (ANNs) using naive neurons while there are types of neurons and other brain cells that have not been well studied and understood. So CNN or even SNN, they are not similar to brains at some degrees.
- How do neurons talk to each other?
The majority of neurons use one of two common signaling molecules known as neurotransmitters, GABA or glutamate, that are known to pass through specialized synapses. But there are many other types of signaling molecules present in the brain, and it’s not clear how those molecules get their message across.
The way neurons interract with others in ANNs is now fairly simple even the AI communtity has been groaning about the complexity of models, so what will happen if the connection changes?
- How does the brain compute?
We are using calculus and algebra to simulate the computation but Koch said:
Neuroscientists have been studying the visual part of the mammalian brain for decades, but until very recently technology only allowed them to capture information from a handful of neurons at a time. It’s like if you tried to watch a movie but could only see 1000 pixels out of several million on the screen.
What does it mean? To me, we are trying to fake the computation while the real picture has not been captured yet.
— Hanoi, Jun of 21 —