Video and audio input/output will be essential parts to a truly human-like AI, but they aren't the core of it. Just giving some computer the ability to move around and see and hear isn't going to give it the complexity of a human brain. YOu have to figure out how to teach it how to use all these peripherals intelligently. That's the hard part the original poster seems to have skipped over.

But what's this fascination with a completel human-like AI? We have humans already. I'm more interested in a computer-like AI. Something that can communicate with us and think logically and solve problems in math and science and maybe a little literature and art as well. I think a machine like this will happen long before a human-like AI comes aorund. Mainly because there's little motivation to make a human-like AI.

We want a machine that can do work in computer science and engineering. In science matters it can have a virtual IQ of 300 and tell us the immediate right direction to go in to make much faster more intelligent machines. After that, ANY kind of AI is trivial because we have a machine that can emulate the productivity of an entire generation of super intelligent humans with unlimited resources.

Quote: Original post by Kelly G
A computer cannot think. It can only compute. It can add, subtract, multiply, and divide. It can tell if two numbers are the same or if one is greater or less than the other. It can move numbers from one device to another. That's really all. Everything else is a combination of these. You describe a series of steps that an automataton can follow but you did not say how it would compute how to do these things, and that is exactly what information is missing and what people are trying to research.

Maybe that's all us humans can do as well? Thought/consciousness and any other 'human like' attributes may simply be a bi-product of a complicated network.

Quote: Original post by Gullanian
Maybe that's all us humans can do as well? Thought/consciousness and any other 'human like' attributes may simply be a bi-product of a complicated network.

The answer to that is not decided either way yet. Till it is known maybe we should hold off on trying to program a human-like AI? Otherwise we might end up developing in the wrong direction. Besides, without knowing what the human brain does, how would we have a baseline to decide if the AI has met the goal or not yet?

in favor of us being more than big calculators:
There is always the 'i think, therefore i think i am' argument...

in favor of us being nothing but big calculators:
Theres the research into people with 'split-brain'
where their heispheres have been surgically separated, and operate independantly yet outwardly they act like a normal person.
This would seem to indicate that either the Cartesian Theater resides in only one hemisphere; or that there is no Theater at all! or that it resides in the combined system and people with this operation are in fact true zombies now.

If the brain is just a NN, then the problem is one of input. Of course this can't be the case. There must be some hardcoded data in the DNA, the goals, self preservation, and so on. Of course there can't be a specific gene for any of this, rather many genes with some complex relation. It may be easier to start from scratch than to try to decypher the DNA.

besides fuzzy logic and the short memory cycle discussed already, humans do not work on a single processor. Our upper conciousness does, but we have innumerably more that change everything we do, memory in itself is so dynamic and is altered every moment that trying to emulate it in a computer (with very stationary memory) is close to impossible. Almost a third of your brain is separate processors that return different charges based on the charge inputted. EVEN THAT is dynamic based on the person, so while one person might approach a basic "fetch" problem by visually scanning the shelves, others might check the index or a computer catalogue, or even asking another human!

Discounting ALL of that, you're looking at something that can sucessfully identify items, people, locations, human logic and dialects, and would basically be a god among machines to be able to do all that in any short amount of time (<1 min).

With our current neurological knowledge and digital computing power, it is impossible.

Not to mention, what if someone stole the book?

It's estimated that at any one time we may have up to 11 million sensory inputs, that's going to be fairly tricky to model on a computer.

Quote: Original post by Anonymous Poster
in favor of us being more than big calculators:
There is always the 'i think, therefore i think i am' argument...

That argument was shot down 150 years ago, but more elegantly refuted by Sartre with his formulation, existence precludes essence. At any rate, consciousness is not computational ... (A Critique of the Computer Theory of Mind)

i think we already know how a human brain is working, but we dont understand it.

see: i'll tell you know how a computer works:
a pc consists of transistors. a transistor is an electronical component that has two inputs and one output. if both inputs are energized, the output is energized too. this is done by a semi-conductor.

ok, thats all, now build a pc. or at least a part of it, a ram or a cpu.

i think you have been at school, so you know about electricity and stuff. but i think you can't build a pc with the information you have. but it can be done, at least you are sitting in front of a pc, don't you ?

the same with the human brain. you have a cell that has some inputs and outputs. when the inputs reach some level of voltage, the outputs are energized. it even works with electricity.

now go and build a brain.
i think, it is very well possible. but we don't know what to do to build it up.

and there is one thing, a machine cannot do: reproduction.
if a human learns something, the cells of the brain build new connections. something a mechanical part can never do by itself, until you build a teleporter or replicator.
or, like the human body, a "framework" where the brain acts in.

All you have to do is simulate a 3d matrix of 10billion units with an average of 10 thousand weighted connections each with each node operating simultaneously.

Just run this pseudocode

long i
For I = 0 to pow(10, 10)
CreateThread(MyNeuron)
Next

Then in MyNeuron just connect to the 10 thousand other neurons try to connect more to nearer neurons. You might want to alter this to simulate inter-axon communication.

You might also want to investigate the 3-4 billion years of evolution to discover the initial clustering, activation and connection rules that apply.

I don't know why I didn't try this back on my Vic20!

Quote: Original post by ROBERTREAD1
All you have to do is simulate a 3d matrix of 10billion units with an average of 10 thousand weighted connections each with each node operating simultaneously.

Not necessarially.
We do not know that Scale is the only defining factor for the human brain; it may very well be that the brain uses a non-Turing model of computation that we have not yet discovered. In which case your proposal is hosed.