oops, I meant to add that those people with math anxiety should have a look at evolutionary techniques of training ANNs. (thats what I was getting at in my first answer mongrel, I should have been more specific - I''m blaming it all on lack of sleep ;0))



Stimulate

I am quite happy with the replies made. First, the closest answer to my main question was from someone who had no clue what he was talking about. Maybe there IS a natural reward/punishment system in the human brain that causes the neurons to reset their synapse weights.

When i said that if we had no senses, we couldnt think, I meant that if we had no senses EVER. Meaning, if we sensed one thing, one time, then we could think...though it would be limited severely.

Someone said that the human brain is not just an I/O machine, that we can set our own conditions. What is meant exactly? And how is abstract thought done in the brain?

Also about the stages of ANN being different from the brain''s continuous I/O process, it would not be too difficult to simulate that coninuity. All you have to do is make a system of cycles where it checks all input and output at a specific point in time.

I plan on looking at the Kohonen self-organizing system. Also, calculus is not a difficult topic for me. I just have not had it yet (I am in 11th grade...will be taking Calculus next year...)

Thanks for the help.

quote: Original post by Taile

Also about the stages of ANN being different from the brain''s continuous I/O process, it would not be too difficult to simulate that coninuity. All you have to do is make a system of cycles where it checks all input and output at a specific point in time.

Not true. You''re assuming that the process is linear and that it can be stopped (or viewed at any time) and analysed and then restarted again. It cannot... the process is exceptionally non-linear and non-Markovian, so looking at it at just one instant and evaluating inputs and outputs doesn''t work. The thing about ANNs (and adaptive Belief networks, which are really the same thing except with completely different internal semantics) is that they are built so as to be Markovian, so that we CAN implement them computationally.

Non-Markovian dynamics are a PIG to compute... and it takes some hideous mathematics based in quantum field theory to describe them!!! One of my good friends did this as his PhD to implement computationally the equations that govern turbulence, something that had never been done before and had only been simulated using direct numerical simulation techniques on simplified problems that were assumed Markovian. His thesis is a masterpiece of mathematics but almost impossible to understand! There are maybe 6 people in the world that can!

This is not to say that we cannot obtain useful results from Markovian approximations... but ultimately they don''t tell us what is going on in the real, non-Markovian system.

Cheers,

Timkin

quote:
unless we have a soul, there is not a goal from which to set the brain. Well, does anyone have any insight into this?

couple things here. you seem to be asking how do i take a blank neural network and design an algorithm such that the neural network starts learning stuff.

ok there are a couple assumptions buried in there that are totally wrong.

First asking to start at nothing and get something implies that you think the brain starts with "reset synapse weights". Through your fetal development your brain is set up to process certain things. that''s why everyone in the world has a Broca''s area and a visual cortex whose architectures are pretty much the same as for everyone else on the planet. there''s no starting at zero. you are effectively preprogrammed by your genetics to learn certain types of things.

Second the whole notion of resetting synapse weights. dunno where that came from but there''s no resetting the brain. it''s a continual process of modification.

So basically the question comes down to what are you trying to accomplish with this network.

do you want human like intelligence?

when you say i want it to learn, what kinds of things do you want it to learn? how to walk? how to talk? remember bits of information? learn to think logically?

the brain is not a general purpose learning mechanism. all of the above things are leart by completely seperate systems that have been evolved to learn specific types of things. though the mechanism of learning seems to be similar for all those components, what each component is capable of learning is totally different.

and you don''t need a soul to learn. just set up a network in a certain way and it''ll process data in a certain way. connect the outputs of that network to the right part of the body and it does the right thing. there''s no master control center evaluating what things should be learned and whatnot. it all just happens systematically building off of that massivly pre-programmed architecture.

my suggestion is that if you want to "simulate a brain" you best be taking some neurobiology courses or at least learning some neurobiology.

-me

you cant even begin to work with human like ai, since you dont seem to have any remote understanding of biology and nature (i almost hit the back button when i ran across your first mention of "soul").

go get some books on evolution. i recommend the richard dawkins line, like, "the blind watch maker", or "climbing mount improbable", and take some biology classes.

"step out of the box", seriously.

You''re an atheist aren''t you? ;P
Well, seeing the word ''soul'' and running away from it is simply a sign of a narrow mind. That is as bad as the Christians panicking at the sight of the word ''evolution''. But no offense is meant. I simply disagree.

I am an agnostic. I don''t necessarily believe in a soul: I have no reason to believe they exist. But I dont have any reason to disbelieve either. I was only suggesting the possibility of a soul when I mentioned it above.

But if we had a soul, then it definitely would have something to do with our thought, and therefore our brain would not be all we think it is. Only a possibility.

Hey guys, discussing about the existence of a soul is leading nowhere in this topic
Just one little thought: if the soul -assuming it exists- is in the heart, for example, what if you have a heart-exchanging operation? *gg* Did the quys in white change your soul?

Well, never mind, was just a little idea that popped into my mind. Ok, and here''s my real contribution: I agree with you in this respect: the wish to build a human-like AI-system really requires an in-depth biology knowledge. As far as I know, science today isn''t even able to explain every single procedure of the brain, so it remains a mysterium for now.
You see, starting work on a system that should work like a human''s brain seems quite useless at the moment...
Maybe we need the knowledge of the variables Einstein mentioned in the past. He argued that everything would be predictable if you just knew some (still missing) values to calculate with.

Well, it''s quite late now, I''m gotta go to sleep, but I''ll come back later, because I''m interested in the development of this thread...let''s see...

Yours,

Indeterminatus

--si tacuisses, philosophus mansisses--

Hey guys, discussing about the existence of a soul is leading nowhere in this topic
Just one little thought: if the soul -assuming it exists- is in the heart, for example, what if you have a heart-exchanging operation? *gg* Did the quys in white change your soul?

Well, never mind, was just a little idea that popped into my mind. Ok, and here''s my real contribution: I agree with you in this respect: the wish to build a human-like AI-system really requires an in-depth biology knowledge. As far as I know, science today isn''t even able to explain every single procedure of the brain, so it remains a mysterium for now.
You see, starting work on a system that should work like a human''s brain seems quite useless at the moment...
Maybe we need the knowledge of the variables Einstein mentioned in the past. He argued that everything would be predictable if you just knew some (still missing) values to calculate with.

Well, it''s quite late now, I''m gotta go to sleep, but I''ll come back later, because I''m interested in the development of this thread...let''s see...

Yours,

Indeterminatus

--si tacuisses, philosophus mansisses--

Hey guys, discussing about the existence of a soul is leading nowhere in this topic
Just one little thought: if the soul -assuming it exists- is in the heart, for example, what if you have a heart-exchanging operation? *gg* Did the quys in white change your soul?

Well, never mind, was just a little idea that popped into my mind. Ok, and here''s my real contribution: I agree with you in this respect: the wish to build a human-like AI-system really requires an in-depth biology knowledge. As far as I know, science today isn''t even able to explain every single procedure of the brain, so it remains a mysterium for now.
You see, starting work on a system that should work like a human''s brain seems quite useless at the moment...
Maybe we need the knowledge of the variables Einstein mentioned in the past. He argued that everything would be predictable if you just knew some (still missing) values to calculate with.

Well, it''s quite late now, I''m gotta go to sleep, but I''ll come back later, because I''m interested in the development of this thread...let''s see...

Yours,

Indeterminatus

--si tacuisses, philosophus mansisses--

A small tip...

Dont listen to Kohonen. Make up your own theories and if you can, skip as much calculus as you possibly can.

Although you need some to get around some major obstacles, or you have to invent new ways of doing stuff, like my team did

Today I am very happy, because our Brain-child spoke for the first time. Articles of how we did it should be out in a couple of years. (Yep, has to be revised and approved first :I )

But our way is just one way of doing stuff, so go along your own path and you might strike lightning from a blue sky.

Right now we would need some industry standard for AI and ANN''s so we can design the very first commercial AI-accelerator cards.

/Dr. H.P. Neumann and ''Hull'' the 1337 coder.