Hacker News Reader

Comment by peter303 | Hacker News Reader

The Brain is like XXXX technology, where XXXX is: 1) Clockwork (18th century) 2) Factory (19th century) 3) Computer (10th century) 4) Nueral Net (21st century) 5) None of the above! Ding Ding Ding

View on HN

It would seem to be increasingly similar to each of those things in turn, not "none of the above".

The brain is quite literally a neural network (in a more traditional sense of the word "neural"). A biological brain (especially a simple one) could probably be modeled somewhat accurately and with reasonable efficiency by a cyclic NN.

Unless you have religious objections, it's also quite reasonable to call the brain a computer.

But how accurate would the simulation be? Real brains have latencies as well as action potential. I don't think it's obvious how memories are stored, as well.

I have no objections to brain being modeled through a bunch of linear algebra. I'm not sure that we have all the particular interactions required for a full model yet?

A model containing only neurons may be too simplistic - glial cells also seem to modulate cognition in ways that haven't yet been fully elucidated. Then you also have rather more asymmetric synaptic behaviours such as volume transmission, where the neurotransmitter is diffused across a less well-defined area, rather than being kept neatly within a shared synaptic cleft.

You are arguing against argument by analogy with an analogy.

In any case none of those analogies are entirely wrong, they are just increasingly more accurate. An eighteenth century person couldn't have understood computers, so clockwork is the closest analogy.

I mean no disrespect, but you are making the same argument that those people did :) I think the more general truth is that we humans explain using analogy, and the brain is complex, so we use something else that isn't fully understood by most.

> but you are making the same argument that those people did

of course that person is. GP is saying "each model is a better approximation, but still an approximation" - and you are saying "they all said that, and they were all wrong."

but are they getting _less_ wrong over time?

Technically they're all the same Turing machine analogy. (Recurrent neural nets aren't really Turing complete, but they become Turing complete if you add repeatable state logic.)

It's just that the later analogies include some hint of emergent properties that are missing from the earlier ones.

Is the brain really a Turing machine? I doubt it's the right model, but obviously I have no way to prove this.

But here's a question - if the brain is a symbolic data processor, what exactly does it process? This sounds like it should have an obvious answer, but the more you think about it (ironically...) the less obvious the answer becomes.

>You are arguing against argument by analogy with an analogy

No, he's arguing against a specific argument by analogy with an analogy.

Or rather, he's not even doing that. He's arguing not against "argument by analogy" (in general) or a "specific argument by analogy" (in particular), but against an analogy bare from argument.

>Factory (19th century)

Specifically, a factory with a steam engine, hence the language of internal "drives" and so on.

The 21c part of this analogy should be "the internet," following the principle of us basing our mind-model on the newest technology around at the time.

I always thought it odd to say that our mind is like a computer, when we created computers in the image of our minds.

>I always thought it odd to say that our mind is like a computer, when we created computers in the image of our minds.

Only we didn't do that. We created computers as machines doing arithmetic. In the image of our abacuses would be a better description.

Can an abacus do arithmetic?

Before we had the mechanical implementations, "computer" was a human occupation.

Except a neural network is much like a simulation of neurons. So its not "like a computer," we're using computers to implement a model based roughly on the biological brain. Its an intentional implementation to mimick the brain.

That's a misconception. A neural net is just a mathematical algorithm, not any kind of simulation of the human brain.

I don't see the conflict here. Simulations are mathematical algorithms. The mathematical algorithms of neural networks were devised as a model of what we observed neurons doing and turned out to be quite good at some of the things brains are quite good at.

This is quite an overstatement, the electric potential of a neuron would be most accurately described by some pde, only approximately by the ode's that are usually used. Then it is another bunch of approximation down the line until you arrive at neural networks. It might be the case that certain types of neural networks accurately model the computational essence of what a brain does, but it does so in quite a different way.

Reply on news.ycombinator.com