Mission to Build A Simulated Brain Begins

Link: http://www.newscient...le.ns?id=dn7470

---

Mission to build a simulated brain begins
00:01 06 June 2005
NewScientist.com news service
Duncan Graham-Rowe

An effort to create the first computer simulation of the entire human brain, right down to the molecular level, was launched on Monday.

The “Blue Brain” project, a collaboration between IBM and a Swiss university team, will involve building a custom-made supercomputer based on IBM’s Blue Gene design.

The hope is that the virtual brain will help shed light on some aspects of human cognition, such as perception, memory and perhaps even consciousness.

It will be the first time humans will be able to observe the electrical code our brains use to represent the world, and to do so in real time, say Henry Markram, director of Brain and Mind Institute at the Ecole Polytecnique Fédérale de Lausanne (EPFL), Switzerland.

It may also help in understanding how certain malfunctions of the brain’s “microcircuits” could cause psychiatric disorders such as autism, schizophrenia and depression, he says.

Until now this sort of undertaking would not be possible because the processing power and the scientific knowledge of how the brain is wired simply was not there, says Charles Peck, IBM’s lead researcher on the project.

“But there has been a convergence of the biological data and the computational resources,” he says. But efforts to map the brain’s circuits and the development of the Blue Gene supercomputer, which has a peak processing power of at least 22.8 teraflops, now make this possible.

Mapping the brain
For over a decade Markram and his colleagues have been building a database of the neural architecture of the neocortex, the largest and most complex part of mammalian brains.

Using pioneering techniques, they have studied precisely how individual neurons behave electrically and built up a set of rules for how different types of neurons connect to one another.

Very thin slices of mouse brain were kept alive under a microscope and probed electrically before being stained to reveal the synaptic, or nerve, connections. “We have the largest database in the world of single neurons that have been recorded and stained,” says Markram.

Neocortical columns
Using this database the initial phase of Blue Brain will model the electrical structure of neocortical columns - neural circuits that are repeated throughout the brain.

“These are the network units of the brain,” says Markram. Measuring just 0.5 millimetres by 2 mm, these units contain between 10 and 70,000 neurons, depending upon the species.

Once this is complete, the behaviour of columns can be mapped and modelled before moving into the second phase of the project.

Two new models will be built, one a molecular model of the neurons involved. The other will clone the behavioural model of columns thousands of times to produce a complete neocortex, and eventually the rest of the brain.

The end product, which will take at least a decade to achieve, can then be stimulated and observed to see how different parts of the brain behave. For example, visual information can be inputted to the visual cortex, while Blue Brain’s response is observed.

Amazing!

Shame the project is about 10 years from completion... but atleast they are starting on it now

I can't wait to hear about the end results of this project and what it means in terms of building up possibly Super Intelligence?

I'm impressed. I wonder if oscrazor believes this is as significant as it seems.

I had no idea this was possible in these times. According to Bill Joy we'll have the CPU power to simulate a single cell in 2010. Assuming his calculations were correct, I never would have thought a project like this would be launched in 2005.

But hey... IBM is behind this. And IBM is the same company behind the revolutionary Cell chip, so I guess there's no reason to not take this very seriously.

I'm pretty sure the world would learn quite a lot from a simulation like this. In fact, I think this might bring the Singularity a few years closer.

Well, read carefully between the lines. The computer in question would be hard pressed to fold a single large protein in anything less than hours, and probably even days (I don't know how fast the current state of the art folding systems can work, but it's nowhere near real-time, by several orders of magnitude).

And a single cell has tens of thousands of proteins that need to be folded at some point or another, possibly even millions or more. So this simulation isn't really going to be at the level of granularity that Bill Joy was talking about when he predicted simulating a cell in 2010.

More than likely, this will simulate much higher level behaviors. Rather than folding a protein, it'll just know that protein A can do P, Q, and R, and takes up dimensions MxNxO, and has mass K, and reactive points at locations X, Y, and Z, etc., etc., and then simulate millions of such proteins together. It may not even be that granular, but use such heuristics as concentrations of various proteins and other macro- and micromolecules, rather than simulating the individual molecules.

This will still be quite fascinating in its own right, but depends on us knowing a lot about the character of each molecule, and having accurate heuristics of what happens at various concentrations, etc. With a full-blown atomic/molecular simulaiton, you just need the basic laws of quantum physics, as applied to the atomic and molecular level, and let the simulation work out the character of each molecule as a result of running the simulation. (Some extra heuristics for common aspects of "chemistry" could be used to get a couple orders of magnitude speed-up, at a small loss of generality.)

But that would require not teraflops, but yottaflops (trillions of teraflops) at the least, and exabytes (millions of terabytes) of RAM, at the least, for a single cell.

Of course, this level of precision isn't absolutely necessary to get mostly correct results.

You can find an audio of a discussion about this project at the bottom of this post.

Next month, IBM is set to activate the most ambitious simulation of a human brain yet conceived. It's a model they say is accurate down to the molecule.

No one claims the "Blue Brain" project will be self-aware. But this project, and others like it, uses electrical patterns in a silicon brain to simulate the electrical patterns in the human brain -- patterns which are intimately linked to thought.

But if computer programs start generating these patterns -- these electrical "thoughts" -- then what separates us from them? Traditionally human beings have reserved words like "reasoning," "self-awareness," and "soul" as their exclusive property. But with the stirring of something akin to electronic consciousness -- some argue that human beings need to give up the ghost, and embrace the machine in all of us.

Marvin Minsky, Brian Cantwell Smith, and Paul Davies discuss this on WBUR radio.

Listen to the show:

http://realserver.bu...06/con_0613b.rm

Thanks for the audio link. I find those to be very helpful, since I put them on CD and listen when I drive.

I wonder if there will be a brain@home project starting in 10 years? It seems to me that many (many!) different areas of the brain will have to be calculated and compiled in order to get a vision of the whole.

Maybe we should all take $2000 and set it aside, and then in 10 years purchase the best computer we can - just to speed this research along (if the brain@home thing happens).

I've looked for an FAQ on this - this whole thing is quite exciting.

Here is their FAQ.

Sadly, the researchers don't yet have a distributed computing project in the works yet. Maybe in a couple years - which means that the 'common man' can't contribute to this research for some time yet. Sigh. This is really breakthrough stuff that I'd like to see advance as quickly as possible.

This is interesting work but it's also fairly useless for immediate use in life extension.

They're approaching this as an attempt to design better AI (They say they're not, but seriously, "a brain on a chip"? Without even meeting them it's pretty easy to tell what's going on subconsciously!) and gain better understanding of how to build a brain, from scratch, at a molecular level. This research would be far better directed if they were to concentrate on how to transfer a living brain's logic to a more reliable form.

I also take issue with the idea that you need to pull the brain apart at a molecular level to be able to understand the emergent conscious state to it's entirity. This is what I have begun to identify as "the scientist's religion", the idea of soul or some level far too complex for us to ever understand or master.

This research is approaching the brain like we used to look at the genetic code for humans and think it was incredibly complex and what made us so unique from other animals. Now we can pull it apart, we can see that 99% of our genetic code also belongs to a banana and that a lot of it is just evolutionary carry over that never gets expressed. Using a bit of forethought and logic, I am going to predict that this research will back up the idea that 99% of the networking and molecular interaction in a human brain is the same as that in a rat. And that will be some major discovery all over Nature.

Great for philosophical discussion of god and genetic inheritance of logic.

Doesn't really help me much when it comes to dying.

It's like saying you need to understand the molecular logic behind bone growth to be able to look at a broken joint and design a replacement. As far as I am aware, the logic of a brain occurs at a bulk molecular level, neurotransmission and ion depolarisation involving thousands of units, operating in bulk, per neuron. Each molecule of a neurotransmitter can be much bigger than the junction of a transistor. A synaptic junction isn't even in the running when compared with a modern day transistor, no matter what parameter of it you're discussing, size, rise and fall time to and from mesa and valley, bandwidth etc. The only difference is that transistors don't reorganise themselves at a hardware level within processor cores, yet. Simple, mechanical answer, have the processor redesign it's own mask for the next generation of the processor based on what it's lacking most often. More complex, add transistors to swich generic arrays of transistors between functions.

Before anyone tries to argue that a synaptic junction is an organic, continuous device, review you understanding of transistors (particularly biasing) and how axons depolarise (in discrete, digital bursts at a frequency of ~1kHz maximum). A transistor can depolarise at GHz quite happily, a few million times faster than an organic axon, and in a highly continuous fashion (Transistors aren't restrained to the on/off digital operation those without electronics experience quote; see audio amplifiers & biasing again).

Seems that almost all attempts to replicate a human brain focus on the idea of just modelling it at a higher and higher resolution, whilst simultaneously missing out all of the other macro features that make it work the way it does (a neural network for empirical observation). That also raises questions over whether or not you'll be creating an intelligence that will be conscious of being imprisoned in a world where it's blind, deaf etc.

Even if the research turns up a cure for brain disease, I can still die of heart, lung, liver, kidney ...insert whatever you like... failure.

The key goal of medicine should be indefinite life extension. Once you have that, you have all the time in the world to explore the ifs and maybes of the universe. The route they should take to this should be gaining the ability to transfer human consciousness into a more reliable form. The human body is majorly flawed and, to this date, has terminally failed on a routine basis without fail. What use is a cure for AIDS or heart disease, or whatever else the body can fail due to, if you already have the ability to transfer a consciousness from organic to synthetic form?

I'm seriously tired with people making elements of the body out to be more than they are. Learn from our past mistakes. When you start assuming something is too complex to understand from the start, you -will- end up kicking yourself, guaranteed.

I am 100% confident that a transistor junction based brain, capable of the same input / output capacity (Today's computers don't have eyes, ears, skin, a nose, ears etc and so rely on us to tell them about their world) and the ability to reorganise it's junction (A logic gain greater >1), could beat a human brain. Easily. If you think otherwise, please join an organised religion and return any implants or drugs / supplements you have.

The research is useful, since any understanding is better than none, but focused excessively in an area of minor importance area in my opinion.

They'll have to kick the millitary guys off their supercomputer simulations of bigger and better nuclear weapons first I guess.

They should've given me the money. I'm sure I could have spent it on a more worthwhile objective. Save up all your $2k's and then give them to me. They've tried the other way for thousand of years now, hygiene / drug therapy / surgery, and gained tens of years of life extension. Now try the John way, transference to a synthetic form, and gain indefinite years! [lol]

Edited by johnuk, 29 November 2005 - 01:14 AM.