Re: AI and Natural language
> I don't really remember much biology, but doesn't a single brain cell
> just conduct an electrical charge down the length of the cell? Think that's
> what they taught us...anyway, a piece of copper wire could emulate that. No
> big computer needed. The other function of a cell being to grow a little
> and then die sometime, a blade of grass could emulate that better than a
> computer. So we plant us an acre full of grass, does that emulate the
> brain?
Neurology lesson, the squeamish may turn off now. 
In its simplest sense, the neuron is more of a switch than a wire. It takes
inputs, sums them up, compares the sum with its internal threshold voltage
(the significance of which will be discussed shortly), and if it surpasses
threshold, fires a signal of its own. Inputs can be inhibitory, where they
will detract from the signal (gamma-alpha-butyric acid I think is an
inhibitory neurotransmitter), or excitatory (noradrenaline), where they add
to it.
The behavior of neurons can be adjusted by changing their threshold. This is
the basis of neural networks. By changing the amount of stimulus needed to
evoke a response, and by wiring a series of neurons together, complex responses
to multiple inputs can be generated.
A "few moments" to simulate a brain cell may refer to the learning time to
train the virtual neuron to a task (train meaning find the optimal threshold
voltage to respond correctly to a task) though a single-layer, single-celled
network is not that useful. Neural network testing I did during my undergrad
days involved a three-tier network with six to eight nodes; this network
was used to do language acquisition testing (simple things like determine the
agent/actor in the sentence, etc.) and compare its progress with human subjects
doing the same task. My undergrad thesis was based on this project. Fun
--
Cameron Kaiser * spectre at sserv.com * http://www.sserv.com/
--
Visit the leading Internet starting point today!
http://www.sserv.com/
--
|
Not Categorized, Please Help
|
|
