Re: Blue Brain: Illuminating the Mind

we thought we would...

Not likely from this particular approach.

I'll be fascinated to see how the connectivity issue with Blue Gene plays out, or whether the issue is even allowed to surface. The rigid compartmentalization and problem layout with nearest-neighbor communication imposed by the Blue Gene architecture sounds wrong for simulating the human brain, but what do I know?

RM

Yes, meant to say 10-million "times".

How do you give a machine delusions? Delusions such as qualia? Sensor qualia such as color, odor, sound, etc.? Self qualia such as personhood, consciousness, awareness, etc.? Religious qualia such as wonder, meaning, conscience, etc.?

-- Best, Frederick Martin McNeill Poway, California, United States of America snipped-for-privacy@fuzzysys.com

************************* Phrase of the week : "Somewhere in this process, you will come face-to-face with the sudden and shocking realization that you are completely crazy. Your mind is a shrieking, gibbering madhouse on wheels barreling pell-mell down the hill, utterly out of control and hopeless. No problem. You are not crazier than you were yesterday. It has always been this way and you never noticed."

- Henepola Gunaratama :-))))Snort!)

*************************

--
Seven orders of magnitude.

     1E11
    ------ = 1E7
     1E4

Does that assure us that the subjective experiences associated with those delusions will be experienced by the machine? I know we primate folk take our delusions for granted, but here we have a sibling machine, wouldn't want to "short change" it.

We want a person type machine, not a zombie type. (Though some known defective humanoids are ersatz persons.)

It looks like we may arrive at true artificial intelligence a lot sooner than we thought we would... Quote from article below: "Markram's EPFL team, collaborating with IBM researchers and an online network of brain and computer scientists, will use Blue Brain to create a detailed computer model of the neocortex, the largest and most complex part of the human brain. "That's going to take two to three years," he says. Then, with a bigger Blue Brain, he hopes to build a cellular-level model of the entire brain. This may take a decade..." Full Article: ; ; BUSINESS WEEK NEWS ANALYSIS ; By Otis Port ; June, 2005 ;

; content/jun2005/tc2005066_6414_tc024.htm ; ; Blue Brain: Illuminating the Mind ; ; Scientists will use the blazingly fast supercomputer to do ; never-before-possible research into how we think and how mental ; disorders arise ; ; On July 1, the Blue Brain computer will wake up, marking "a ; monumental moment" in the history of brain research, says ; neuroscientist Henry Markram, founder of the Brain Mind ; Institute at Switzerland's Ecole Polytechnique Federale de ; Lausanne (EPFL). The event could usher in a new era of ; scientific discoveries about the workings of the human mind. ; ; The Blue Brain computer is the latest installation of IBM's ; (IBM ) BlueGene/L system, a radically new approach in ; supercomputer design. EPFL's machine has a peak speed of some ; 22.8 teraflops -- meaning it can theoretically spit out 22.8 ; trillion calculations every second. That blazing speed should ; put Blue Brain among the world's top 15 supercomputers. (The ; world champ is the BlueGene system at Lawrence Livermore ; National Laboratory -- when finished later this year, it will ; have a peak speed of 367 teraflops.) ; ; ; A UNIQUE FACILITY ; Markram's EPFL team, collaborating with IBM researchers and an ; online network of brain and computer scientists, will use Blue ; Brain to create a detailed computer model of the neocortex, the ; largest and most complex part of the human brain. "That's going ; to take two to three years," he says. Then, with a bigger ; Blue Brain, he hopes to build a cellular-level model of the ; entire brain. This may take a decade -- even with IBM's ; next-generation system, BlueGene/P. Markram can't wait to get ; his hands on one of these number-crunching beasts. ; ; BlueGene/P will have faster processors and could ultimately ; reach petaflops speeds-- quadrillions of calculations per ; second. "We're planning on a very long-term effort," notes ; Markram. "We're creating a unique facility for researchers ; worldwide." Adds Charles Peck, the IBM researcher who leads ; the Blue Brain effort at IBM's research division in Yorktown ; Heights, N.Y.: "There's now a tremendous opportunity to do ; some science that up to this point just hasn't been possible." ; ; THINKING MYSTERY ; The Blue Brain Project will search for novel insights into how ; humans think and remember. Plus, by by running accurate ; simulations of brain processes, "we'll be able to investigate ; questions about psychiatric disorders and how they arise," ; Markram says. Scientists believe that autism, schizophrenia, ; depression, and other psychological problems are caused by ; defective or malfunctioning circuitry in the brain. ; ; Parkinson's disease is another target, adds Markram. "There's a ; group of cells deep down in the mid-brain that produce ; dopamine, and when these cells begin to die and dopamine ; production decreases, you get Parkinson's," he explains. "We'll ; be able to mimic this," creating simulations that should make ; Blue Brain an invaluable tool for drug-company researchers on ; the track of treatments or cures for Parkinson's. ; ; Learning how the brain works has been one of science's great ; challenges. Researchers still don't have a holistic grasp of ; how we think. One reason: Most research so far has been ; conducted with "wet" experiments -- stimulating or dissecting ; the brains of mice, rats, and other animals. Markram notes that ; "some 'wet-lab' experiments are incredibly complicated," taking ; up to three years and costing $1 million. ; ; With simulations on Blue Brain, he predicts, "we'll be able to ; do that same work in days, maybe seconds. It's going to be ; absolutely phenomenal." ; ; CONSTANTLY CHANGING CIRCUITRY ; Markram first broached the idea of a BlueGene-based ; collaboration five years ago, right after IBM unveiled the ; supercomputer system. "Even before that, Henry had been wanting ; to go down this path of computer simulations," says IBM's Peck. ; "But only now is it actually feasible." ; ; That's because the brain is so extraordinarily complex that an ; enormously powerful computer is required. The brain's physical ; structure and electrochemical operations are very intricate. ; Complicating things still further is its constantly changing ; internal circuitry. "The brain is in a very different state in ; the morning, when you wake up, than it is at noontime," Markram ; points out. ; ; Fifty years ago, he notes, "we believed that memories were ; somehow hardwired into the brain. But our lab [EPFL's ; Laboratory of Neural Microcircuitry] has been one of the main ; propagators of a new theory, in which the brain is incredibly ; fluid. It's restructuring itself continuously -- ; self-organizing and reorganizing all the time." ; ; HUGE SIMULATION ; If brain circuitry is in a constant state of flux, Markram ; insists that long-term memories can't be permanent, hardwired ; fixtures. To explain how memories are preserved, he and his ; colleagues cooked up the "liquid-computing" theory. Validating ; this concept with Blue Brain, he hints, might point to new ; types of silicon circuits that perform new and more-complex ; functions -- which IBM could use to build a revolutionary ; brain-like computer. ; ; "That's a possibility," says Tilak Agerwala, a vice-president ; at IBM Research. "But we're still very far from understanding ; how the brain works, so it's much too early to know if we ; should build computers that way." However, the notion already ; has a fancy moniker: biometaphorical computing. ; ; For now, Markram sees the BlueGene architecture as the best ; tool for modeling the brain. Blue Brain has some 8,000 ; processors, and by mapping one or two simulated brain neurons ; to each processor, the computer will become a silicon replica ; of 10,000 neurons. "Then we'll interconnect them with the rules ; [in software] that we've worked out about how the brain ; functions," says Markram. ; ; The result will be a full-fledged model of 10,000 neurons ; jabbering back and forth -- a simulation 1,000 times larger ; than any similar model to date. ; ; FANTASTIC ACCELERATION ; This setup will form the foundation for studying neocortical ; columns -- the building blocks of the cortex and the part of ; the brain that differentiates mammals from other animals. Each ; column is a bundle of networked neurons and is roughly 1/2 ; millimeter in diameter and 2 millimeters long. That's only ; about the size of a pinhead, Markram notes. "But packed inside ; are 50,000 neurons and more than 5 kilometers [3 miles] of ; wiring," he marvels. ; ; "The neocortical column is the beginning of intelligence and ; adaptability," Markram adds. "It marks the jump from reptiles ; to mammals." When it evolved, it was like Mother Nature had ; discovered the Pentium chip, he quips -- the circuitry "was so ; successful that it's just duplicated, with very little ; variation, from mouse to man. In the human cortex, there are ; just more cortical columns -- about 1 million." ; ; Since the neocortical column was first discovered 40 years ago, ; researchers have been painstakingly unraveling how it helps ; perform the miracles of thought that enable humans to be ; creative, inventive, philosophical creatures. "That's been my ; passion, my mission for 10 years," says Markram. "Now, we know ; how information is transferred form one neuron to another. We ; know how they behave -- what they do and whom they talk to. ; We've actually mapped that out." ; ; Next, that knowledge will be transferred into a torridly fast ; silicon simulator. Blue Brain promises a fantastic acceleration ; in brain research. It could be as dramatic as the leap from ; chiseling numbers in Sumerian clay tablets 2,500 years ago to ; crunching them in modern computers. And the Blue Brain Project ; just might culminate in a new breed of supersmart computers ; that will make even BlueGene/L seem like a piker. ; ; Otis Port is a senior writer for BusinessWeek in New York ;

The way I see it is that technology always tries to model itself after nature. I've noticed that computer systems and the internet works very similarly to my old workplace which was the library. They're called information sciences for a reason, I guess.

But models will be models and I don't really think it would be able to

*replace* humans. If anything, we'll come up with something that might be better than us in certain ways, but it'll just be different, not better.

That's true. Fortunately our models of what nature is up to are constantly being refined. At this time our culture supports abysmally anachronistic models of brain structure and function. We are stuck similarly as the medical community was stuck with the miasma model of disease or the engineering community was stuck with the phlogiston model of heat.

The machines probably won't *replace* us, they will be good siblings and friends. The machines will help us produce improved genetic modifications that will compete with the *natural* model.

When (and if) we encounter intelligent extraterrestrials we probably will learn truly different ways, perhaps better. A putative ET race that has been around for a billion years should have something to show for it.

As a side note, I suspect that computers (as such) days are numbered. The problem is I/O. They won't be able to compete with information systems that attach directly to our own 'wetware'. However, once that happens, who is to say where our own consciousness ends, and the 'computer' begins? This is already happening, in a crude form. Many people use Google as a kind of memory... once we can simply 'remember' things on Google, as needed, what *are* we?

My view is that at that point, *we* will be the AIs. There won't be a need for clumsy separate entities, with their own desires. We will provide the 'intentionality', 'it' will provide the communication, memory and computation.

If the computational and biological facilities become sufficiently advanced, there is no telling what will happen to us, where 'we' will end, and 'it' will begin. A whole new chapter.

--
Regards,
  Bob Monsen

we thought we would...

Agreed.

Your cortex uses two-level wiring: There's short-distance wiring using non-myelinated fibes. Due to the delays introduced - signal speed is less than a meter per second - these are at most a few tens of millimeters long, and most are significantly shorter. The long distance wiring uses myelinated fibres and is what you will find in your brain atlas as the white matter. (The grey matter are the neurons mixed with the short-distance wiring - a layer only a few millimeters thick on the surface of the cortex.) While their speed is up to 100 m/s, there is a much smaller number of them - I remember something 3 orders of magnitude less; nonetheless, due to the bulk of the myelination, they actually occupy most of the cortex's volume.

Of course, the brain has the advantage of using three-dimensional, self- connecting and adaptive wiring.

Jan

A short answer would say, you bind the machine's state, whatever that be, to the information it processes.

Content-Transfer-Encoding: 8Bit

Jan Vorbrüggen wrote:

So that's about 12 years if Moore's "Law" holds...

It's one less - cortex is only about 10B neurons, the other 90% are mostly in the cerebellum doing God knows what.

Jan

One tiny column is 50.000 neurons, so this will only be 20%....

As usually technically uneducated journalists get overly excited about something they don't have a clue about. I hope it helps raise funds, at least.

--
Siol
------------------------------------------------
Rather than a heartless beep
Or a rude error message,
See these simple words: "File not found."

You say we take our delusions for granted, but I wonder to what extent we enter into agreements on what we experience, only accepting as delusions those things for which there is no agreement.

As much as we share very similar hardware, our experiences are communicated by agreements. A machine would need to lean our agreements to communicate its experience.

Leaving aside that aspect of imagination which creates its own agreements, I wonder if communication, internal and external, is the same as experience? Do we think in terms of language / vision? When i say i love, or i feel pain, do we mean the same thing? Can i assume experience, from the language used to describe it? .

Put another way, a machine might see the color red, as a particular frequency of light, its qualia of that experience, might be everything else it associates with that color. Bound to its internal state, as various systems are made to act upon that experience, it would form a qualia of a sorts. Could we accept that? Would that be the same as our experience of the color? I doubt we would be satisfied, since we know we have evolved bound to the many advantages of seeing the color red. Yet it would be experience.

BTw I share your goal for this machine, yet i wonder where the person is found, in the person. Is it found in the difference between people, the differences between man and the rest of nature? Or is it found in what man can claim as common? Could we ever view a machine as person-like, and accept where it differs, or would be reject the attempt, since from the start we know it will always be different. Even when the machine exceeds our capacity for thought, it would always be classed a non-person.

I suspect they aren't worring about the archtecture difference at this point. They probably just want access to a lot of computing power to run larger simulations to see how they behave.

I suspsect there are still some very important "parts" they don't unerstand yet and when they try to simulate what they do know, all they will find is that it's not doing anything "intelligent", which only tells them they are still missing something important.

I think it's important to try these simulations to help lead to better understanding of the parts, but I don't really expect it to make the type of progress they seem to be hoping for. They really seem to be talking as if the _only_ thing missing is a big enough machine to run the simulation, and when they get that, they will have a human brain simulation creating human behavior.

That would only be true if they correctly understood what the brain was doing. I don't believe they understand that yet. And when they build a supercomputer sized simulation of the wrong low level behavior, all they will get out of it is billions of numbers per second of garbage. Only when they get the full and complete picture of what is happening at the low level will the simulation do something interesting.

But maybe they understand more than I realize?

--
Curt Welch                                            http://CurtWelch.Com/
curt@kcwc.com                                        http://NewsReader.Com/

we thought we would...

There used to be talk about doing a full computer models of an ant/cockroach/flatworm/something-else-with-a-small-brain that would do what the actual organism does in all situations. There were even claims (with fuzzy details) that it had already been done. *Has* such a simulation ever been done, and if so how well did it work?

The quality of the simulation is dubious at best. From the physiological perspective, to fully simulate a single neuron in terms of all the relevant processes involved in receptor up and down regulation, protein synthesis, gene expression, autocrine, paracrine and endocrine responses, etc., all of which are vital to excitability, it might take the entire computer to do even a remotely decent simulation of one neuron, and that would assume we knew a lot more than we do now about all these processes.

If one wanted to draw a crude analogy between such a computer and a living animal brain, one might be tempted to equate one CPU with one neuron. In that case, one would have to equate the biologically important components such as amino acids and nucleotides to the transistors that make up the CPU. While it would be possible to multitask the simulation since not all of these things change in outwardly physiologically significant ways at nanosecond intervals, the number of transistors required would far exceed any known fabrication process, because the brain is an analog device, so modeling it with a digital device requires vastly more logic gates.

There is little doubt that useful information will emerge which may lead to productive developments, but the notion that we are even remotely close to being able to actually replace any significant "wet lab" experiment, even on one cell, with any computer is hyperbole.

Basically, what this simulation may permit, depending once again on the quality of the implementation, is some greater degree of insight into how networks of neurons might interact and permit exploration of simpler experimental variables to study the effects of cell loss in degenerative diseases, for instance.

OTOH, if the goal is to model consciousness, most of the efforts in that direction have been quite abstract and departed largely from the microanatomy and physiology of the human brain. Rather, they have been modeled on more abstract observations of how the brain appears to behave in response to environmental stimuli.

We now call that, roughly speaking, "entropy." Still magical stuff. :-)