Contrasting a human brain to a computer

[His Divine Shadow]

Please, lets have your input on this issue.

A computer is essentially fast but dumb
A brain is slow but smart.

Thats the dumbed down definition.

The Computer is a binary processor and works in states of and off

The Human brain is a massively paralleled neural network, it's individual nerouns have on and off states, but they also possess infinite states of excitement inbetween, essentially, they're analouge, this is the reason for our intelligence and adaptability.


Now the question is, how would be best design a sentient machine? Prefferably one with the pros of a machine and a human.

-Would it be with billions of very small on-off "nodes" linked togehter like in a neural network, essentially imitating a human brain, but in a digital fashion.

-Or would it be through a massively fast and capable supercomputer with some extremely advanced software?

-Also, would you say that the assertion that a human brain is analouge correct?

[Defiant]

-Would it be with billions of very small on-off "nodes" linked togehter like in a neural network, essentially imitating a human brain, but in a digital fashion.

-Or would it be through a massively fast and capable supercomputer with some extremely advanced software?

-Also, would you say that the assertion that a human brain is analouge correct?

I would go with the first choice. Physical devices that operate in a similar fasion to neurons but are much faster is probably doable in the foreseeable future. If you go with the second choice, you would need incredibly complex software, and its been my experience that the more complex software is, the greater the chance of something stupid going wrong with it (just ask M$).

As for the analog assumption, what are you basing that on?

[His Divine Shadow]

As for the analog assumption, what are you basing that on?

Well there are more than on/off states for neurons, there are also levels of excitement, thats analouge then I believe.

[Arrow]

If you want a machine that is self-aware and able to learn, you need a massive parallel network. The feasibility of this design comes down to the complexity and size of the network (and increase in one general corresponds to a decrease in the other). The software that each node in the network executes should be fairly simple; it only needs to model the states and transitions found in one neuron. Neural nets are amazing things; a friend once wrote a 1000 node neural net that was able to learn the grammar of a simple language based on examples of its usage.

Now if you want a machine with an answer to everything, your going to have to lose the self-aware bit. In this design you'll comparitively few processes than what the above requires. Rather, you need a very well designed database able to parse the questions given to it and return the results quickly. Google is a primitive example of this.

[His Divine Shadow]

I never specified such a thing.

Anyway, there seems to me that one can accomplish this neural network in two fashions.

1) Physical neural network with individual physical neurons of cell size
About 100 billion of these and we ought to have a human equivalent brain, probably superior to a human since in this, signals travel much faster and there are probably other plusses, anyone care to elaborate?

2) Megacomputer, like a Cray machine that runs these neural networks as software.

[His Divine Shadow]

This made me think of possible cybernetic upgrades to fleshy human brains, through the use of nano-technology.

1) Nanobots in the brain examine neurons and their connections, creates an electronic duplicate and replaces it, and repeats the process, this might be a process that takes years, you'd probably not notice it happening either, this could be a recipie for immortality.

2) Partial replacement, possibly just replacing the "wires"(axons) between neurons with inorganic ones that send information at speeds like that in a computer.
One could also possibly create artificial synapses in addition and skip the whole chemical conversion part.

[Arrow]

I never specified such a thing.

Anyway, there seems to me that one can accomplish this neural network in two fashions.

1) Physical neural network with individual physical neurons of cell size
About 100 billion of these and we ought to have a human equivalent brain, probably superior to a human since in this, signals travel much faster and there are probably other plusses, anyone care to elaborate?

2) Megacomputer, like a Cray machine that runs these neural networks as software.

Option two probably won't work. You want option one, as each node can run all the time. If you use option two, the OS has to unload some nodes in order to allow others to execute; it would result in switching back and forth (which is what Window's does - take a look at the processes in the Task Manager - it unloads one process from the CPU and gives another the CPU for execution millions of times a second). The result of switching all of the nodes would slow down the AI. Now, if nanotech gives us some ultra-fast processors or useful quantum computers are constructed, then option two could work, but at added software complexity (because now you need a robust OS).

[His Divine Shadow]

Hmm, assuming such hyper-tech, what about entire nano-computers that can each do many many operations and use them as neurons, hmm, that'd be quite fast, no?

[NecronLord]

It's a software problem, not hardware, we can make machines that keep the heart and other systems running now, but we can't create the software to run higher order functions.

[His Divine Shadow]

It's a software problem, not hardware, we can make machines that keep the heart and other systems running now, but we can't create the software to run higher order functions.

Actually, it's hardware too, since neural networking is the only concievable way to get true sentient AI's and the best way of doing it is to make neuron sized artifical ones.

What you are essentially talking about is option #2 here.

[Dahak]

The Human brain is a massively paralleled neural network, it's individual nerouns have on and off states, but they also possess infinite states of excitement inbetween, essentially, they're analouge, this is the reason for our intelligence and adaptability.

Neurons don't have infinite states of excitement. They exactly have one.

You either reach the excitement level which makes the neuron fire, or you don't.
But a neuron can, up to a certain limit, increase the frequency of firing off pulses.

[His Divine Shadow]

Well, is the human brain analouge or digital or what then?

[NecronLord]

It's a software problem, not hardware, we can make machines that keep the heart and other systems running now, but we can't create the software to run higher order functions.

Actually, it's hardware too, since neural networking is the only concievable way to get true sentient AI's and the best way of doing it is to make neuron sized artifical ones.

What you are essentially talking about is option #2 here.

Not the best way, the most advanced way. The problem is maintainance, with option 2 maintainance becomes fairly easy, as every few minutes the AI is backed up to non-volitile storage, with option 1 a good kick screws it.

[NecronLord]

Well, is the human brain analouge or digital or what then?

It's... organic. We don't need to make an exact copy of the human brain, just copy the good bits and drop the redundancies.

[His Divine Shadow]

Not the best way, the most advanced way. The problem is maintainance, with option 2 maintainance becomes fairly easy, as every few minutes the AI is backed up to non-volitile storage, with option 1 a good kick screws it.

Exactly whats your proof of that? A human head can take a kicking and come out of it without problems.

And the tech level of such a brain would require advanced micro-tech on the cellular scale, "repair/maintenance-cells" is not impossible.
And this "computer", or rather, it's neurons and axons would likely be suspended in some fluid of some form and be quite "mushy" or jello like, not hard like a CPU or circuit board.

[UltraViolence83]

Sounds like the bio-gel from Voyager.

[His Divine Shadow]

Sounds like the bio-gel from Voyager.

Nah, thats just a good concept spoiled, they do use neural systems in a suspended gel like pack, but they're biological.

[UltraViolence83]

Hmm...How can an electronic machine be "mushy" if it isn't biological? I mean, eletricity doesn't take kindly to wet things.

[Kenny_10_Bellys]

I think you're making a basic error here, it may be necessary at first to divorce the content from the system containing it. The computational theory of mind explains the way the mind works using modern computer software design (and to a lesser extent the hardware) to explain how the mind copes with the amazing problems it overcomes on a day to day basis.

It theorises that the mind is made up of 'demons' or 'agents' that take care of very specific tasks, essentially custom designed modules for performing very specific tasks and passing them on to other modules for further work. It's reckoned that your vision system has somewhere around 30+ modules or agents processing everything you see, trying to make sense of it all, finding edges, recognising people, working out shape and 3D, etc. Each module does this by breaking a hugely complex task down into easier tasks, then break down again and again until you reach tasks so easy that a very simple device can handle them, perhaps even just digital on-off switches. So it would be with a computer.

If you can design these basic modules (not easy) your problem then becomes one of parallel networking the modules together, which could possibly be acheived by networking many small computers together, but I doubt a mainframe large enough and simple enough could be built for the task anytime soon. But if you have the software or firmware to hand that will do the simple thinking, then it could almost become platform independant and the modules could be made to fit each successive refinement of technology that shrinks the size of the computational devices used, until perhaps we reach the level of cybernetics perhaps.

Computer and robotics designers are working right now on the basics of vision and touch systems and it's painfully obvious how difficult a task it really is to design modules to do these apparently simple tasks, but obviously not impossible. With concerted effort, you never know what might be round the next corner. But first you need to design those modules!

[His Divine Shadow]

Hmm...How can an electronic machine be "mushy" if it isn't biological? I mean, eletricity doesn't take kindly to wet things.

It would be nigh impossible for it not to be mushy if it was made like a human brain, just because it's not organic doesn't mean it has to be hard.

There are non-conducting fluids.
Eletricity is in our mushy biological systems too you know.

And when you think of it, a bunch of cell sized artificial neurons in a human like eletronic brain with similar dimensions would leave lots of empty space, hence some kinda gel to suspend it in.

[His Divine Shadow]

But first you need to design those modules!

This is not about creating interfaces, not even close, this is about AI, the "software" of the brain is the interaction of the neurons, what we're talking about is duplicating the human brain through artificial equivalents.

We don't need any software when the human brain is our template.

[Admiral Valdemar]

Why do we have to make our own? Why not just mass produce brains and have them work together, eventually they'll learn much like the rat did which was plugged into a computer.

[His Divine Shadow]

Why do we have to make our own? Why not just mass produce brains and have them work together, eventually they'll learn much like the rat did which was plugged into a computer.

Make our own what?

What rat?

[Kenny_10_Bellys]

Again you appear to be confusing two different things. You want intelligence, not matter. Can you take apart a motorcycle and extract a bucket of speed? The brain is certainly an amazing device, but the intelligence is not inherent in the axions themselves, although they are necessary for running the brains software. If it was, is a small lump of brain matter slightly intelligent? Cells in a jar? Are chimps intelligent? Their brains are incredibly similar to ours, but they're not considered intelligent. You could say that about almost all mammals, same type of brain material but they're not intelligent.

The stuff i explained above (rather badly) is detailed in Professor Steven Pinkers book 'How the mind works' and the computational theory is pretty much the widely accepted view of how psychology and evolution came up with our intelligence. You are a collection of software subroutines running on a marvellous organic computer, but you are a product of the software more than anything.

[His Divine Shadow]

Again you appear to be confusing two different things. You want intelligence, not matter. Can you take apart a motorcycle and extract a bucket of speed?

You appear to be under some serious misconception on how the brain works, it does not have "software", it's equivalent of software is directly linked to the neurons interacting with each other, the physical and eletrical are merged.

To simplify it, you seem to think the brain has programs, that are like stored on a hardrive or something, they don't, they are neurons interacting and making connections, change the neurons connections and it'll become something else.

It's not like stored on a harddrive and you can say delete it without altering the brain in a physical manner.