3 ms·

I'm always wary of thinking that our metaphors for the brain represent the real object. History shows that we tend to use as metaphor whatever popular technolo

I'm always wary of thinking that our metaphors for the brain represent the real object. History shows that we tend to use as metaphor whatever popular technology is in place at the time. Right now is neural networks. Before, computers, telegraphs, hydraulic systems and so on. I could easily imagine the next analogy being quantum computers. Or whatever trending technology comes around 2030.

godelski 3y ago
Metaphors are aids. The problem is to take metaphors literally. It is, ironically, over-fitting the meaning of words. But words only mean what we mean them to mean. In the language of ML, you can say that -- analogously or metaphorically -- language is kinda like a VAE. You encode your thoughts, put them into a compressed version that is then passed to another independent network which must decode this encoding and hope you got it all right. While far from accurate, it is useful and can help us remember that the goal is not to interpret the words, but what was attempted to be encoded in the first place. That said, a computer is a very vague term. I think if you strictly mean the device we are typing to one another on or the thing that transmits the signals between our machines, you have overfit. Even with how the word itself has historically been used as well as even used today. A person can be called a computer as well as a physical slide ruler. It is simply a machine that can calculate, which is absolutely another vague term. In fact, if the universe is completely mechanistic, we could frame it as a computer (but do not confuse this with all that simulation stuff, that's a misunderstanding of what a universe as a simulation actually means). FWIW, Roger Penrose for quite some time been making the connection of a brain to a quantum computer. The problem with that is we aren't exactly seeing quantum processes in the way we'd expect (though there are clearly some quantum processes). It gets nuanced, but it's a rabbit hole that sounds like you'll enjoy going down.

godelski 3y ago

Metaphors are aids. The problem is to take metaphors literally. It is, ironically, over-fitting the meaning of words. But words only mean what we mean them to mean. In the language of ML, you can say that -- analogously or metaphorically -- language is kinda like a VAE. You encode your thoughts, put them into a compressed version that is then passed to another independent network which must decode this encoding and hope you got it all right. While far from accurate, it is useful and can help us remember that the goal is not to interpret the words, but what was attempted to be encoded in the first place. That said, a computer is a very vague term. I think if you strictly mean the device we are typing to one another on or the thing that transmits the signals between our machines, you have overfit. Even with how the word itself has historically been used as well as even used today. A person can be called a computer as well as a physical slide ruler. It is simply a machine that can calculate, which is absolutely another vague term. In fact, if the universe is completely mechanistic, we could frame it as a computer (but do not confuse this with all that simulation stuff, that's a misunderstanding of what a universe as a simulation actually means). FWIW, Roger Penrose for quite some time been making the connection of a brain to a quantum computer. The problem with that is we aren't exactly seeing quantum processes in the way we'd expect (though there are clearly some quantum processes). It gets nuanced, but it's a rabbit hole that sounds like you'll enjoy going down.