Artificial human brain 10 years away

[Rob060870]

Hello readers, I have been thinking and if you bear with me hopefully I don’t sound too mad!.

I have read an interesting article and in it a leading scientist has claimed that A detailed, functional artificial human brain can be built within the next 10 years. Henry Markram, director of the Blue Brain Project, has already simulated elements of a rat brain.

Lets go further, if this is true then what if these artificial human brains were put into virtual bodies’ living on a virtual planet in yes you have guessed it, a virtual Galaxy and universe. Now these people would think just like us and could live out their lives just like we do in their own virtual environment and world.
A virtual human could live a whole lifetime but this could take place in just a few milliseconds for us, these people would have families, loved ones, inventers, great thinkers etc. they would evolve very quickly and in time would surpass our own evolution of technology and biology.

Would it not be exciting to copy their technology and use it for our own purposes?
They may invent better super computers, time travel, or faster than light space travel or master immortality! Even simulate virtual people of their own! And so on and so forth.. I really do believe that this deserves serious thought.

Maybe we ARE virtual people ourselves running this experiment for others!
Is this possible?, What do you think?


Read more: http://realitypod.com/2010/03/artificial-brain/#ixzz15SAxkeXO

http://www.ted.com/talks/view/id/659

* source realitypod.com

 

[NeoDevin]

Sounds like a good videogame, as long as you can fast forward the boring parts.

 

[Mech_Engineer]

Such a prediction is probably being made based on the number of transistors expected to be in computer processors in 10 years. Let's just say having the same number of transistors in a computer chip as cells in a human brain is still a long ways from simulating a human brain, and farther still from that is simulating millions or billions of brains interacting in a simulated world.

Still your idea is an interesting one, until a brain becomes Skynet ;)

 

[turbo]

Human brains build and maintain (and lose) pathways, and reform themselves throughout our lives. Raw computing power cannot replicate that. Perhaps someday, but not 10 years from now.

 

[Mech_Engineer]

Human brains build and maintain (and lose) pathways, and reform themselves throughout our lives. Raw computing power cannot replicate that. Perhaps someday, but not 10 years from now.

We're closer than you might think, adaptive hardware using genetic algorithms and self-healing neural networks can achieve that right now, albeit on a much smaller scale.

 

[waht]

Our brains have about 10^11 neurons, and more importantly, 10^15 connections between them (from wiki). As a result the brain is an extremely massive parallel system. Currently, the fastest computer in the world is running at 1.5 petaflop or 10^15 floating point operations per second derived exclusively from massive parallel computing. Not that this is a fair comparison, but it gives you idea where we're at, the brain is different in other ways, but we are catching up fast.

And it will happen within our lifetimes. It's just a matter of deductive logic. Consider the following projection:

By 2020 we are expecting a 10^18 flop machine. That's 3 orders of magnitudes faster than what's currently available. Perhaps by 2030 it will be 10^25?

http://img101.imageshack.us/img101/8576/exaflopcomputers.jpg

 

[Mech_Engineer]

I would argue that the predictive line between 2010 and 2020 isn't steep enough based on the apparent slope of the curve thus far!

 

[waht]

I would argue that the predictive line between 2010 and 2020 isn't steep enough based on the apparent slope of the curve thus far!

It's a ballpark estimate. If you look at growth from 2000 to 2010, the slope suddenly jumped, but then it leveled a bit.

 

[stevenb]

A detailed, functional artificial human brain can be built within the next 10 years.

If there exist detailed, functional real human brains that are dumb enough to believe this statement, then it's probably true.

 

[hamster143]

I would argue that the predictive line between 2010 and 2020 isn't steep enough based on the apparent slope of the curve thus far!

Not so fast.

Supercomputers spent the last 10 years "catching up" to the industry. The fastest supercomputer of 2001 was running on 250/350 nm, 375 MHz processors. By the time it was commissioned, Intel was actively selling 180 nm 1.7 GHz processors and preparing to launch its 130 nm line. It was rated around 7 teraflops. If you were to build a system with the same number of contemporary Pentium 4's, it would rate an order of magnitude faster and it would produce less waste heat, to boot.

Fast forward to 2010: Tianhe-1A. Lag time from the release of the underlying chip (Tesla M2050) to the release of the supercomputer: less than 6 months.

So what do we have in the end? In nine years, fabrication technology progresses from 180 nm to 40 nm. Power consumption per transistor drops accordingly, as a square of transistor size: by about a factor of 20. Clock speeds hit the brick wall.

At this rate of progression, by 2019 we should have mass-production 10 nm GPGPUs. (Some expect to see first 11 nm chips in 2015, but that's probably too optimistic. Intel might demonstrate 11 nm in 2016 if all goes well. GPGPUs will be a year or two behind.) That will take us from 2.5 petaflops to 40 petaflops. In another 6-8 years ,or in 2025-2027, assuming that photolitography does not break down below 10 nm or it is replaced by a suitable alternative (and fast!), we'll get close to the 10^18 mark.

 

[DaveC426913]

Lets go further, if this is true then what if these artificial human brains were put into virtual bodies’ living on a virtual planet in yes you have guessed it, a virtual Galaxy and universe. Now these people would think just like us and could live out their lives just like we do in their own virtual environment and world.

How did you get from 'artificial brain' to 'virtual body on a virtual planet'?

'Artificial' and 'virtual' are apples and oranges.

And what causes you to assume that an artifical brain would be even the slightest bit like a human brain? A car made out of wood is still a car, but I would not count on it operating anything like a Honda Civic.

 

[JaredJames]

I agree with DaveC on this one.

It's one thing to replicate the processing power of the human brain, but a whole different issue generating the consciousness (and other human traits) of a person. They're not in the same country, let alone ball park.

 

[Rob060870]

How did you get from 'artificial brain' to 'virtual body on a virtual planet'?

'Artificial' and 'virtual' are apples and oranges.

And what causes you to assume that an artifical brain would be even the slightest bit like a human brain? A car made out of wood is still a car, but I would not count on it operating anything like a Honda Civic.

I am not an expert in this matter I am only asking the question.
Interesting point however,

Is an artificial human brain not a virtual one? as it exists inside a computer?
why wouldn’t a detailed, functional artificial human brain with a similar architecture, the 50-100 billion neurons and 1000 trillion synaptic connections act in a similar way to a biological brain?

and if you gave this brain a virtual body and environment to explore could it not learn from stimuli?

My thinking was rightly or wrongly that if a human brain is made successfully inside a computer then to give it stimuli to help it learn just as we do, it could be placed inside a virtual body and environment as well. This would help this brain to learn just like our brains do. The environment that this artificial intelligent being was placed in would, hopefully in time be as realistic as the world that we live in today. Obeying the same laws as physics as us.



"In June 2005, Henry Markram, director of the Blue Brain project, announced his intention to build a human brain using one of the most powerful supercomputers in the world. "The critics were unbelievable," recalls Markram. "Everybody thought we were crazy. Even the most eminent computational neuroscientists and theoreticians said the project would fail."
Some of Markram's peers said there simply wasn't enough data available to simulate a human brain. "There is no neuroscientist on the planet that has the authority to say we don't understand enough," says Markram. "We all know a tiny slice. Nobody even knows how much we know."
Markram was not dissuaded by the negative reaction to his announcement. Two years on, he has already developed a computer simulation of the neocortical column - the basic building block of the neocortex, the higher functioning part of our brains - of a two-week-old rat, and it behaves exactly like its biological counterpart. It's something quite beautiful when you watch it pulse on the giant 3D screens the researchers have constructed.
The neocortical column is the most recently evolved part of our brain and is responsible for such things as reasoning and self-awareness. It was a quantum leap in evolution. The human brain contains a thousand times more neocortical columns than a rat's brain, but there is very little difference, biologically speaking, between a rat's brain and our own. Build one column, and you can effectively build the entire neocortex - if you have the computational power.
Although a neocortical column is only 2 millimetres long and half a millimetre in diameter, it contains 10,000 neurons and 30m synapses. The machine that simulates this column is an IBM Blue Gene/L supercomputer is capable of speeds of 18.7 trillion calculations per second. It has 8,000 processors and is one of the most powerful supercomputers in the world.
Markram believes that with the state of technology today, it is possible to build an entire rat's neocortex, which is the next phase of the Blue Brain project, due to begin next year. From there, it's cats, then monkeys and finally, a human brain."

(Source/ the guardian)


Full link- http:www.guardian.co.uk/technology/2007/dec/20/research.it

 

[Mech_Engineer]

Is an artificial human brain not a virtual one? as it exists inside a computer?
why wouldn’t a detailed, functional artificial human brain with a similar architecture, the 50-100 billion neurons and 1000 trillion synaptic connections act in a similar way to a biological brain?

Well the problem is the brain isn't just a bunch of neurons, there are learning patterns and instincts built into our DNA which would have to be described in the simulation as well. Replicating the number of neurons is easy compared to that...

and if you gave this brain a virtual body and environment to explore could it not learn from stimuli?

Yes it theoretically could, and in a simple way that's what artificial neural networks do now.

My thinking was rightly or wrongly that if a human brain is made successfully inside a computer then to give it stimuli to help it learn just as we do, it could be placed inside a virtual body and environment as well. This would help this brain to learn just like our brains do. The environment that this artificial intelligent being was placed in would, hopefully in time be as realistic as the world that we live in today. Obeying the same laws as physics as us.

Well making an environment a brain can learn from is one thing, replicating the human experience in virtual world is a whole lot more work. Making a world that is "just as realistic as our own" would be a huge leap in simulation technology; but of course so would simulating a brain.

 

[DaveC426913]

I am not an expert in this matter I am only asking the question.

Wasn't questioning your expertise, mine is little better, simply examining the logical leaps.

Is an artificial human brain not a virtual one? as it exists inside a computer?

Does it exist inside a computer? Or is it the computer?

I did not read the whole article but from what I have read, the hardware is dedicated to the task of simulating the brain. In order for this brain to have 50 billion neurons, it doesn't simply simulate them in the computer - it would be far too slow - the circuits are the neurons.

So, a computerized brain device is very different from a brain simulated within the cyperspace of a computer. One is a real object, the other is a simulated virtual object.

Regardless, consider this: whether or not the brain itself is real or virtual, its interaction is with the real world, not with any virtual world. i.e. there's no Matrix here.

Which means this:

and if you gave this brain a virtual body and environment to explore could it not learn from stimuli?

Is a red herring. There's no virtual environment.

My thinking was rightly or wrongly that if a human brain is made successfully inside a computer then to give it stimuli to help it learn just as we do, it could be placed inside a virtual body and environment as well. This would help this brain to learn just like our brains do. The environment that this artificial intelligent being was placed in

This virtual thing is a completely different animal.

You are confusing
- artificial: (real, physical (though not biological), interacting with the real world)
with
- virtual: (non-physical, simulated within a cyberspace, interacting with a virtual, cyberspatial environment).

 

[waht]

Our consciousness is completely transparent to the machine it's riding on. We are completely unaware of the internal processes that are going in our brains.

In fact, our self-awareness plays a much smaller part in our lives than we care to think. Most of the activities we are doing are autonomous feedback patterns that are being played out, like walking upright, riding a bike, or reaching for a glass of milk. You just call that subroutine, and it magically is executed without you knowing all the lower level details about it.

So in theory it would be possible one day to simulate an authentic human experience on a different hardware. Will it be entirely human? I don't think so. But who is to say our consciousness experience is the only one possible. In fact there might be infinite possibilities of different conscious experiences, and more intelligent at that.

We don't have stop at 10^15? What would intelligence be like at 10^20?

 

[Negatron]

A detailed, functional artificial human brain can be built within the next 10 years.

The slight problem with this, if I recall correctly, is that Markram is trying to applying evolutionary biology to go from rat to human to avoid the vast issue of obtaining a human brain blueprint, which is a much bigger problem due to it's relative size. If things go to plan it could potentially be human-like, but it is a fair assumptions that countless human subtleties will be beyond it. Automated neural scanning systems will have to advance vastly if we are to obtain artificial human analogs.

A virtual human could live a whole lifetime but this could take place in just a few milliseconds for us

Unlikely given we don't even have the processing power to simulate a human lifetime in a lifetime. Conceptually sound but the practical constraints make this a very distant possibility.

Perhaps an equally interesting alternative is to simulate a human "lifetime" over billions of years. The virtual human would experience the universe changing in front of their eyes. It is a very plausible means of time travel if their computer can remain operational, but one way trip only.

Would it not be exciting to copy their technology and use it for our own purposes?

It would be more exciting to simply be them. If their capability and rate of progress vastly exceeds our own it would be much more convenient to simply move in with them rather than linger around in biological obsolescence.

master immortality

There is no inherent reason why a virtual being must age, so immortality would be implied, unless of course we intentionally implemented an aging feature; which would be a bit barbaric. Perhaps you're visualizing a virtual reality which is equally granular to our own, in which case aging would be inherent to the physics of the virtual universe, as it is here, but we have no realistic means of computing such a thing and no way to envision if we ever will so I'm assuming a virtual reality on a much higher level of abstraction which can be computed with physically plausible future hardware.

I really do believe that this deserves serious thought.

It sure does, but hardware and neuroscience must advance much further before it will receive serious thought.

Maybe we ARE virtual people ourselves running this experiment for others!
Is this possible?, What do you think?

Seems unlikely. It would be unlike any conceivable computing device. If true, their Windows install must boot in a picosecond (SSDs no longer seem that fast).

 

[DaveC426913]

Our consciousness is completely transparent to the machine it's riding on. We are completely unaware of the internal processes that are going in our brains.

Sorry, there's no basis for claiming this. With nothing to compare to, we have no way of what is knowing what is transparent. As an easy example: I like to think of my moods as a product of my mind, but they might just as easily be (and quite likely are) a product of varying brain chemistry i.e. a product of my brain's machinery. From my mind's point of view, what is the difference? My moods have always been there, who's to say they're entirely "mine", and not an effect of the machinery my mind runs on?

This means that none of my thoughts can be assumed as coming from my mind as opposed to from my brain machinery.

Unlikely given we don't even have the processing power to simulate a human lifetime in a lifetime. Conceptually sound but the practical constraints make this a very distant possibility.

An excellent point. in terms of processing power, brains are waaay faster than than machines. A machine simulating a brain might take hours to process a single thought. i.e. Rather than a virtual-brain's day occurring next to a real-brain's millisecond, a virtual-brain's day might be more like real brain's month. (That's why we have server farms. There are some things at which computers are really slow.)

 

[Mech_Engineer]

An excellent point. in terms of processing power, brains are waaay faster than than machines. A machine simulating a brain might take hours to process a single thought. i.e. Rather than a virtual-brain's day occurring next to a real-brain's millisecond, a virtual-brain's day might be more like real brain's month. (That's why we have server farms. There are some things at which computers are really slow.)

Well if the artificial brain was constructed like a real brain in terms of neural network connections in a massively parallel processing unit, the processing power and speed should be the same. Rather than making an artificial brain in software where each neuron has to be solved sequentially by a single processing unit, it would be set up more like an artificial neural network with weighted connections in a parallel processing unit (which is why the comparison of number of transistors vs. number of neurons was brought up). As an artificial neural network, as long as the weighting and training is designed to mimic the real thing they should function similarly.

 

[Negatron]

in terms of processing power, brains are waaay faster than than machines.

It's hard to say. If the operating frequency of the neuron was as high as a CPU, there wouldn't be enough power in the world to run a brain full of them. A transistor is much more efficient than a neuron clock for clock, but a better comparison would be a neuron versus a collection of transistors that perform the same function. Assuming a Hodgkin-Huxley neuron the computer is now far less power efficient, however this is largely a result of the high clocks. Reducing clock speed and increasing hardware quantity the computer could potentially match efficiency, although at a much higher cost. This is important because faster doesn't mean a thing without context of the quantity of machines/power required to match.

If a modern GPU can simulate several million HHs in real-time using <400W system power then it would roughly suggest that human scales would be achieved in less than 20 years on desktop hardware, and this is assuming peak activity in all neurons at once which is a bit of a stretch. Brains still have a leg up on modern engineering and they should enjoy their fame while they can; I don't suspect that will be the case for much longer.

 

[Rob060870]

All of this thought is giving my Brain a headache!

 

[DaveC426913]

All of this thought is giving my Brain a headache!

That's a matter of regular maintenance - ensure you change your brain oil every 3000 thoughts or 3 months.

 

[PhilKravitz]

At this rate of progression, by 2019 we should have mass-production 10 nm GPGPUs. (Some expect to see first 11 nm chips in 2015, but that's probably too optimistic. Intel might demonstrate 11 nm in 2016 if all goes well. GPGPUs will be a year or two behind.) That will take us from 2.5 petaflops to 40 petaflops. In another 6-8 years ,or in 2025-2027, assuming that photolitography does not break down below 10 nm or it is replaced by a suitable alternative (and fast!), we'll get close to the 10^18 mark.

This sounds more reasonable to me.

Humans have 1E15 synaptic connections that get evaluated 1E3 times per second so I would say the metric is more like 1E18 for humans.

There is also the bandwidth issue. Not sure how to characterize this for human nor for computers. If you have a super computer spread out over 1000 square feet of floor space yes it may make 1E18 floating point calculation total across the whole floor but they sure as heck are not all talking to each other in a single clock cycle or even 1000s of clock cycles. Whereas the human brain is highly connected in one or two clock cycles.

 

[Pattonias]

It seems to me that the OS of this brain would be the harder problem. The hardware is essentially inevitable. The OS to run it in the fashion that our brain operates would be a hurdle that would be difficult to tackle.

Would the first goal be to create a computer that can generate a truly original idea? Is that when you know that you have arrived?

(I'm not trying to break this down into whether or not original thought exist.)

 

[hamster143]

It seems to me that the OS of this brain would be the harder problem. The hardware is essentially inevitable. The OS to run it in the fashion that our brain operates would be a hurdle that would be difficult to tackle.

If we knew how to design the OS, we wouldn't be really having this conversation right now. Human brain is extremely inefficient, and needing 10^18 flops to simulate it is serious overkill. Consider that a big part of it goes towards image processing, a task that can be performed quite well in real-time with desktop systems doing O(10^10) flops, and that the brain is thought to have only on the order of 1 gigabyte of long-term memory.

But since we can't quite tackle the task of designing the OS (and not for the lack of trying), we're reduced to hoping to circumvent the process, by building an exact replica of an actual human brain as part of the simulation. That way we don't have to figure out all the fine details of deconstructing and modelling someone's thoughts and memories.

Would the first goal be to create a computer that can generate a truly original idea? Is that when you know that you have arrived?

That's way too abstract. There are much more down-to-earth, boring, gritty tasks that seem equally hard to solve.

Start with image recognition. Build a system that can be shown a bunch of pictures of different cars, and given names of their makes and models, and then be able to identify makes and models of other cars (possibly of a different color, with dirt, wear, from a different angle, with lighting variations, perspective distortion, etc.) But be sure not to hard-code any algorithms that are specific to cars. Your system should automatically define, pick and memorize distinguishing features of each model, and then use them to recognize new inputs. This way the system will be adaptable to recognizing boats, buildings, or human faces. The ability to learn would be nice, too.

Last I checked, the preferred approach in this field was, essentially, to give up trying to find a "true" artificial algorithm, and, rather, to create a black box called a "neural network" with a few million interconnected primitive internal nodes. You put a picture in on one end, the answer comes out on the other. You tweak the system till it gives you the right answers. Not very different from trying to build a simulation of the brain (just on a smaller scale).

Then move on to speech. Here things get even more fun, because sometimes speech (sounds) can't be interpreted properly without understanding possible structures behind sounds. For a certain sequence of sounds, there could be 20 different ways to translate it into words, but 15 will be gibberish, 4 will be valid language, and only 1 will have the correct meaning intended by the speaker. So you need to have context, and you need real world knowledge (e.g. that hares can jump, that heirs can inherit property, but hair can't do either). In essence, speech is what separates us from the animals. Even a mouse with the brain that weighs less than a gram can recognize and distinguish objects. Larger animals can memorize names for objects, but only humans and possibly some higher apes have the capacity to grasp the full structure of the language.

Once you have a system that can recognize objects, understand speech, form memories, adding "thought" to the back-end is a relatively trivial task.

 

[Pattonias]

I wonder what it take to have to separate computer systems next to each other with several different systems for communicating, but no inherent(designed) knowledge (programming) of how to use these systems, exchange data. You create an algorithm that's purpose is to receive conformation of the exchange of a particular idea while expressing and idea of its own. It randomly generates data in an attempt at expressing that idea until it manages to "accidentally" use some means of communication that it has at its disposal. The other machine is trying to do the same thing. The experiment is concluded when both machines have exchanged their data with each-other and confirmed an accurate exchange.

You could, given enough time, create a unique form of communication by seeing the means by which the two systems have managed to communicate and by using what tools they two machines had available.

This, to me, would represent a form of simulated brain activity.

 

[hamster143]

I wonder what it take to have to separate computer systems next to each other with several different systems for communicating, but no inherent(designed) knowledge (programming) of how to use these systems, exchange data. You create an algorithm that's purpose is to receive conformation of the exchange of a particular idea while expressing and idea of its own. It randomly generates data in an attempt at expressing that idea until it manages to "accidentally" use some means of communication that it has at its disposal. The other machine is trying to do the same thing. The experiment is concluded when both machines have exchanged their data with each-other and confirmed an accurate exchange.

You could, given enough time, create a unique form of communication by seeing the means by which the two systems have managed to communicate and by using what tools they two machines had available.

This, to me, would represent a form of simulated brain activity.

That's going to take a very long time indeed.

It took O(10^10) years and I'd guess O(10^10) biological entities on the planet at any given time to evolve consciousness. If we want to see same within our lifetimes, we should do better.