Artificial Intelligence!

[Tugrul]

 

 

Hello!

 

Here is a topic that maybe hot on our tails in about 50 years, AI.

 

Okay first for AI to come to life we need full knowlege of the human brain, this includes emotions, thought, institinct and so on. 2nd we need enough computing power to code the complexity of the human brain (we would start small e.g mouse) in a "robot".

 

Now i will discuss how the second aspect could be plausable in 50 year. We are now developing quantum computers, these computers are at their developing stage. About a few years ago 7 atoms figures at 5x3=15. Now that might not seem much but we are only talking about 7 atoms seen as there are millions of atoms in a mm we are talking about billions of terrabytes of computing. This is only re enforced by Moore's Law.

 

Moore's Law is long-term trend in the history of computing hardware, in which the number of transistors that can be placed on an integrated circuit has doubled approximately every two years. Rather than being a naturally-occurring "law" that cannot be controlled, however, Moore's Law is effectively a business practice in which the advancement of transistor counts occurs at a fixed rate.

 

The capabilities of many digital electronic devices are strongly linked to Moore's law: processing speed, memory capacity, sensors and even the number and size of pixels in digital cameras. All of these are improving at exponential rates as well. This has dramatically increased the usefulness of digital electronics in nearly every segment of the world economy.Moore's law precisely describes a driving force of technological and social change in the late 20th and early 21st centuries.

 

Now for the 1st acpect that can make this plausable. Neuroscience has already identified many neurons, and the parts of the brain that code for certain aspects of our being. It is only a matter of time before we completely map the Human brain.

 

Thank you and please discuss

 

Edited November 8, 2009 by Tugrul

[RyanJ]

Shouldn't this be in the Computer Science section? It's more of a computer thing than an engineering thing would you not agree?

 

There is more at the work in the brain than raw processing power. It's more akin to thousands of CPU's working in parallel than one CPU working alone. it's a neural net and even the most advanced ones I've seen only have a hundred "neurons". To emulate the brain, which has tens of billions, is still well outside our understanding.

[Zolar V]

Shouldn't this be in the Computer Science section? It's more of a computer thing than an engineering thing would you not agree?

I think what he is asking is an engineering problem not a computer science.

What i assertain from the post, he is wondering how should we start exploring AI and how it is going to be Engineered

 

To emulate the brain, which has tens of billions, is still well outside our understanding.

i think your wrong, it is fully in our scope of understanding that the brain can be mapped and understood easily within the next few years. The only reason why we have not mapped the brain yet is because of the current laws and past laws making it unlawfull to do many of the things needed to map a human brain. + its underfunded.

[RyanJ]

i think your wrong, it is fully in our scope of understanding that the brain can be mapped and understood easily within the next few years. The only reason why we have not mapped the brain yet is because of the current laws and past laws making it unlawfull to do many of the things needed to map a human brain. + its underfunded.

 

Science is full of things that they said would be "solved in a few years" and most of those are still around today. AI is one of those things.

 

Understanding the brain and mapping it are two completely different things. One can map a genome but it doesn't mean you would understand it. Mapping can be considered the first stage on the path, not the last.

 

And if you read what I said, I said that few neural networks surpass 100 nodes (neurons) and until they do we probably aren't going to make a whole lot of progress based upon what I have seen so far.

[ydoaPs]

Science is full of things that they said would be "solved in a few years" and most of those are still around today. AI is one of those things.

 

Understanding the brain and mapping it are two completely different things. One can map a genome but it doesn't mean you would understand it. Mapping can be considered the first stage on the path, not the last.

 

And if you read what I said, I said that few neural networks surpass 100 nodes (neurons) and until they do we probably aren't going to make a whole lot of progress based upon what I have seen so far.

 

You don't need to map the brain, just make software that

like VitaminD.

[RyanJ]

You don't need to map the brain, just make software that

like VitaminD.

 

I was responding to the post by Zolar V. I never said that it would be a requirement, only that it would help point us in the right direction (probably!).

[bascule]

You don't need to map the brain, just make software that

like VitaminD.

 

We do still need a proper understanding of the structure of the brain. Jeff Hawkins is well aware of this and the role of structures like the thalamus and the loops between the cortex and the thalamus in conscious thought.

[ydoaPs]

We do still need a proper understanding of the structure of the brain. Jeff Hawkins is well aware of this and the role of structures like the thalamus and the loops between the cortex and the thalamus in conscious thought.

 

I wasn't implying that we don't need to study the brain. I was just saying that a neuron-by-neuron map would be essentially useless for AI.

[Zolar V]

understanding the brain could be accomplished in a few years provided that it had adequate funding and a lot of researchers and processing power. just like how we discover what genes do what in the genome project, we could map the complexities of the human brain by testing each part(large scale) one by one. then once we have a rudimentary understanding of the large scale parts go into each part of the brain and test how the parts on the inside of the brain react to each other, then of course test how they interact with other parts.

 

in essence its much like designing a program,building,essay, you start with your superstructure then build in substructures.

[RyanJ]

understanding the brain could be accomplished in a few years provided that it had adequate funding and a lot of researchers and processing power. just like how we discover what genes do what in the genome project, we could map the complexities of the human brain by testing each part(large scale) one by one. then once we have a rudimentary understanding of the large scale parts go into each part of the brain and test how the parts on the inside of the brain react to each other, then of course test how they interact with other parts.

 

in essence its much like designing a program,building,essay, you start with your superstructure then build in substructures.

 

As I said. I'm not convinced. Based upon the many (and I mean many) articles, reviews and so on that I've read - we are a long, long way off understanding the workings of the brain. We've barely begun to scratch the surface.

 

If you wanted to exactly model the brain you would have to create a neural net (or a virtual one) that copies it. And it's never been done, not even an approximation as of yet. It's not a matter of finding - it's the basic matter that you can't emulate something you don't understand.

 

Sure, we may be able to piece bits from here and there and say "xxx" does this while "yyy" does that but we're no closer to understanding how the thing works as a whole than we were 30 years ago.

[CharonY]

From what I have read we are quite farther than 30 years ago. At least in terms of finding out what we still do not understand.

[bascule]

As I said. I'm not convinced. Based upon the many (and I mean many) articles, reviews and so on that I've read - we are a long, long way off understanding the workings of the brain. We've barely begun to scratch the surface.

 

We're well on our way to understanding the brain's high level functions:

 

http://en.wikibooks.org/wiki/Consciousness_Studies/The_Neuroscience_Of_Consciousness

[RyanJ]

From what I have read we are quite farther than 30 years ago. At least in terms of finding out what we still do not understand.

 

At identifying that which we do not understand, maybe. Are we any closer to a complete picture of how the brain works as a whole? Not really. Almost every week there is an article in New Scientist about "x" and "y" feature being found in the brain. It's pretty fascinating really.

[ydoaPs]

At identifying that which we do not understand, maybe. Are we any closer to a complete picture of how the brain works as a whole? Not really. Almost every week there is an article in New Scientist about "x" and "y" feature being found in the brain. It's pretty fascinating really.

 

Do we need to know how the entire brain works? I'm not so sure we do.

[RyanJ]

Do we need to know how the entire brain works? I'm not so sure we do.

 

Honestly? I don't know. I suspect that the brain acts together more than separately so maybe we would. Still, the bottom-up approach of trying to reproduce features from scratch may work - but it may not be an AI in the sense that we think of it.

[ydoaPs]

Honestly? I don't know. I suspect that the brain acts together more than separately so maybe we would. Still, the bottom-up approach of trying to reproduce features from scratch may work - but it may not be an AI in the sense that we think of it.

 

I don't really see the point of human-like AI. IMO, general intelligence is enough. We don't need machines with emotions(see Terminator, Battlestar Galactica, and Matrix as well as several others).

[Tugrul]

I don't really see the point of human-like AI. IMO, general intelligence is enough. We don't need machines with emotions(see Terminator, Battlestar Galactica, and Matrix as well as several others).

 

I would do it just to see that if we can do it.

[bascule]

Do we need to know how the entire brain works? I'm not so sure we do.

 

Much of the brain is unnecessary to conscious thought. The cortex, thalamus, and hippocampus are the three most important structures in conscious thought, although there's a host of ancillary structures we also need to understand (amygdala, locus coeruleus, etc)

[RyanJ]

I don't really see the point of human-like AI. IMO, general intelligence is enough. We don't need machines with emotions(see Terminator, Battlestar Galactica, and Matrix as well as several others).

 

A very good point. But there is no knowing if any type of intelligence would eventually come to act the same way. If it can think for it's self and act for it's self.

[bascule]

I don't think we're going to get very far with strong AI unless we crib directly from biology. So far attempts at guessing the mind's behavior and implementing it in software have failed.

[RyanJ]

Also true. But that's not to say that they will not make some progress in the future. If one could develop a self-modifying neural network that could expand upon information it absorbed then it could eventually become aware. I say could because there is no evidence so far.

[bascule]

Also true. But that's not to say that they will not make some progress in the future. If one could develop a self-modifying neural network that could expand upon information it absorbed then it could eventually become aware. I say could because there is no evidence so far.

 

To put it bluntly, we haven't made any significant advancements in neural networks in the past 30 years.

 

The only kind of neural network I see having future potential is a very different kind of neural network: one which directly tries to model human neural structures. In that regard there is only one project in the entire world attempting this, and that's BlueBrain. Note that BlueBrain is significantly different from what is ordinarily referred to in CS as a neural network.

 

Aside from direct brain emulation attempts like BlueBrain, neural networks are a dead end.

[Zolar V]

you know when i originally said we could map and understand the brain in a few years if we were not constrained by ethics or budget/maning, this is exactly what i meant, "followsreverse-engineer the mammalian brain, in order to understand brain function and dysfunction through detailed simulations. " FROM PROJECT BLUE BRAIN.

now taking their rat tests and replacing it with humans via no ethical restraints, we would significantly reduce the time factor. +if we had gross funding and maning, we would have already made ai imho

[D H]

To put it bluntly, we haven't made any significant advancements in neural networks in the past 30 years. ... Aside from direct brain emulation attempts like BlueBrain, neural networks are a dead end.

???

 

I would argue the other way around: We haven't made any significant advances in symbolic AI in the past 30 years. It is GOFAI, not computational intelligence, that is at a dead end.

[bascule]

I would argue the other way around: We haven't made any significant advances in symbolic AI in the past 30 years. It is GOFAI, not computational intelligence, that is at a dead end.

 

I'm talking about artificial neural networks, not computational intelligence in general.

 

I am excited about the prospects of hierarchical temporal memory.