Re: Blue Brain: Illuminating the Mind

I guess they are aiming for a very accurate realtime simulation of a chunk of neurons about the same size as they can acheive in the lab by wet chemistry. It will be interesting to see how well the refined computer model can emulate actual behaviour.

I would be more interested to see a rough and ready simulation of a much larger chunk of neurons and interconnects not necessarily in realtime.

Astronomy simulations of gravitating particles have just recently crossed the 10^10 barrier eg.

Regards, Martin Brown

Correct assessment, I believe. The proper abstractions have not yet been found.

Jan

....the problem being that it is currently not possible to _measure_ actual behaviour in a way that it could be usefully related to the model's behaviour.

Jan

Do you really believe this? I'm not a sci-fi type, but I suspect that machines that become smarter than their masters are a completely worn-out sci-fi genre, just like predictions of an imminent breakthrough in AI are by now a completely worn-out genre.

That said, AI is one of the places in computation where something really exciting could happen, and really at just about any time, and it needn't have anything to do with exotic new computers; in fact, I rather think it won't. That's because mostly what seems to have been accomplished so far in AI is the realization that no one has a clue. As soon as someone gets a clue, maybe something really interesting will happen. Until that first important clue arrives, I wouldn't be too worried about machine gods. It's so sixties.

RM

"J.Random.User" skrev i en meddelelse news: snipped-for-privacy@corp.supernews.com...

Is this good?

Eventually, some jerk is going to succeed in brute-forcing AI on some quantum computer - without really knowing what he is dealing with - A thing that learns at an exponential rate, is vastly smarter than himself and very much faster, maybe smart and fast enough to learn and predict every thought that the creators could possibly have..

A Machine God.

!Great ;-p

Then ...

Maybe that is why the Universe is so damn quiet - every intelligent lifeform eventually creates something smarter than themselves and loose their intelligence because all answers and needs will be provided for them by the machines they made.

There has been plenty of work on self-organizing systems. I haven't heard anyone who's actually worked in the field say, "If only we had a bigger or faster computer, it might work." Maybe the someone to say that is out there and I can elicit that comment from someone who knows what he's talking about from this post.

The right idea often turns out to be stunningly simple once you have the right idea. So far, no one seems to have stumbled onto the right idea, and building big computers to see if they will turn themselves into a thinking machine by random fiddling doesn't seem like a promising way to go.

Have you read, say, Richard Rhodes on the making of the atomic bomb? A significant amount of basic science by some very bright people went into designing it, and the those laying the money bets had a pretty good idea that it would work and why. Do you have any similar confidence for any scheme for artificial intelligence?

RM

skrev i en meddelelse news: snipped-for-privacy@o13g2000cwo.googlegroups.com...

Even when something is not in vogue it still happens - like a religious war f.ex. One civilisation must destroy another to survive. Such a 12'th century concept, is it not?

Too much top down design, planning and flawed modelling is what kills "AI" I.M.O. Real brains are chaotic, adaptive and evolved - not carefully designed and planned.

The Point is:

Someone might not actually *need* a clue for something really interesting to happen. Maybe general principles and lots of processing power is enough. Maybe Intelligence is really simple to build, simply a property of certain kinds of connected networks that we have not stumbled upon yet or build large enough.

Like handing some U-235 to a medieval cannon-ball maker, "oh, them new cannon balls are heavy and hard, just what we need - so let's make a Really Big One" ;-)

Biological science so far seems to be telling us that Nature worked this way in evolving intelligent life, but perhaps we haven't found the right idea yet. Or perhaps there isn't one.

There have been many people over the years who felt that they had the right approch but just needed a biger or faster version for it to be "intelligent".

Most people however are wiser than that. If it's not acting intelligent on the small scale, making it biger is never going to get us anywhere.

I think I've got a stunningly simple approach that will get us there. So far, everyone I show it to seems to believe it's too simple to be anything new or different or useful.

Whether my network design is going to do anything useful is yet to be seen, but I do strongly believe that intelligence is simply a strong reinforcement learning machine, and that when we do discover how to build this, it will be stunningly simple in comparison to the type of complexity people have been playing with for so long in the search for AI.

Then you would have to listen to it complain about how all the diodes on it's left have begun to ache...

Biological Science takes the above statement on blind faith. I think they haven't got a clue about how it happened, and invoking "Nature" as if it were an entity when it is nothing and doesn't exist, is only covering up that fact. What is this "Nature"? Anything like Gaia?

One would think scientists would be a little more honest.

del cecchi

Wouldn't it be cool to have AI in a AVR, you'd just connect a mic and tell it:

Howdy there partner, would you flash that LED once a second for me?

"SioL" schreef in bericht news:1f%qe.12667$ snipped-for-privacy@news.siol.net...

Ask Eliza what would happen if you did:

"Nature" is simply what is, not some conscious anthropomorphized entity. The Darwinian theory of evolution is largely thought to have involved random mutations, the most adapted of which survived to reproduce and spread their traits.

With respect to AI, there seem to be two main camps. One wants to emulate the behavior of intelligent beings, the other wants to emulate the biophysics. It's not clear that either approach will result in any huge breakthroughs in understanding consciousness.

There was some promise in the careful study of the effects of certain substances on perception and consciousness, but this research was largely suppressed for political reasons around

Well, that does mean that the nodes will spend some of their time passing messages along, instead of just simulating neurons, and, of course, each node will simulate more than one neuron anyways.

It is indeed a non-optimal architecture, but the benefits of a more flexible interconnection layout would be small, since a computer with a flexible interconnect scheme still wouldn't allow a single node as many distant interconnections as thousands of neurons might have.

If this architecture allows a more economical layout with very large numbers of processors, that likely will be more important.

John Savard

The safe statement here is: there's lots of exciting science left to be done.

The field could be divided into those who are trying to identify intelligence as emergent behavior and those who are trying to invent algorithms that mimic intelligent behavior.

The class of those who are looking for intelligence as emergent behavior is really much larger than those trying to do biophysics (or biochemistry, or neuroanatomy or whatever). Identifying intelligence as an emergent behavior of any system would be an exciting breakthrough whether the simulated system had anything to do with natural systems or not.

As distinct from building supercomputers and hoping that they will turn into intelligent beings, the study of emergent behaviors seems to be a serious activity that is worthwhile in its own right. If anyone showed that size matters in the sense that a bigger system could display behavior that is qualitatively different from what a smaller system could display, such a result, in itself, would be fascinating; but the current level of understanding wouldn't seem to warrant the use of gigantic computers.

It isn't enough that we've danced on the edge of the volcano in discussing evolution and intelligence? Now you want to drag consciousness into the discussion? Consciousness is an entirely separate puzzle. I don't think anyone really knows what consciousness is, never mind how to get something like a computer to be conscious.

Oh, piffle. Some professors at Harvard were having fun with drugs. The politics, the mythology, the recreational aspects of the story are all interesting, but they have nothing to do with computers, artificial intelligence, or even, in my humble opinion, the search for the origins of consciousness.

RM

I thought there was something qualitatively different about the neocortex.

The system needs an environment in which to learn to cope, goals required for coping, and a way of interacting with the environment that allows for experimentation and learning.

Physical objects introduce the possibility of introducing something you didn't think to introduce intentionally.

I think Boston's public radio station is set to discuss the Blue Brain silliness today at 11:00 am EDT on The Connection. I heard the announcement, but

RM

By chance there was a piece on the radio this morning about how the songs of bird species are more heavily influenced by location than by genetics: that's how we get a western something-or-other and and eastern something-or-other, same plumage, characteristically different song.

Maybe not only brain size, but learned behavior, the pace of brain development and the length of the child-rearing period (incredibly long for humans) and other environmental factors are important. If that's so, and machine intelligence is to come from mimicking natural processes, we're going to be waiting a long time for such a program to come to fruition. Not only do you need a *very* big computer, and, as you point out, a way of interacting with its environment to learn, you'd also have to provide it with a great deal of "nurturing."

RM

processes, we're going to be waiting a long time for such a program to come to fruition.

.... and if learning to think just is the process of neural interconnection established by training then we won't learn anything much by mimicking a static body of neurons.