You said: 3. Simulation to confirm mathematical models. We are now in the very early phases and call it NN. Later on when more is known, this will morph into something that looks more like present-day ideas as to what AGI will look like, when we start "simulating" the necessary computations and NOT the irrelevant biological details, which we now can't even identify.
I don't see that we fit in this category. I think you can agree it's the high-level behavior of the system that we're trying to replicate in AGI. The question is, how far down do we have to reach to capture that? I don't think we need to simulate all the way down to the neuronal level to build something intelligent. I think the neurons themselves are irrelevant biological details. They are only one possible implementation among many. On Wed, Aug 29, 2012 at 12:52 PM, Steve Richfield <[email protected] > wrote: > Ben, > > I broadly agree with what you have said. I just want to add "a little > amplification" on some points: > > On Wed, Aug 29, 2012 at 8:09 AM, Ben Goertzel <[email protected]> wrote: > >> >> Arets etc., >> >> We do know a lot about the brain now ... but the gaps are so big that >> it's still certainly possible we're missing something fundamental.... The >> roles of astrocytes and glia >> > > Note in passing that ~90% of your brain cells are glial cells. Hence, I > think it is safe to say that getting a handle on the remaining ~10% (as > neuroscience is not doing) might still leave a LOT of unanswered questions. > > >> are still largely unknown; the hints Hameroff keeps pointing out about >> macroscopic quantum phenomena can't wholly be ignored (though I have no >> idea how important they are), along with his (somewhat independent) notion >> of a web of dendro-dendritic connections guiding the flow of attention..... >> And then there's our near utter lack of understanding of large-scale >> complex attractor patterns in the brain, in spite of great simulation work >> like that of Izhikevich & Edelman... >> > > Don't forget the whole area of self-organization, which in my own mind is > the BIGGEST single missing piece. > >> >> For a gov't contract, I recently looked into computational neuroscience >> models of one very particular brain function (accumulation of evidence), >> and the situation is that there are half dozen possible models, mostly >> based on perceptual decision making in monkeys, and nobody has any clear >> idea which model applicable to which parts of the brain, or how accurate >> each of the models is beyond some very simple situations.... Knowledge is >> progressing steadily -- and quickly on the historical scale -- but at the >> current rate it will be >> > > many > > >> decades before the neuroscientists really understand what's going on. >> Though a massive breakthrough is always possible ;) >> > > I think what your are writing may be different than what I am reading. If > there were some sort of single incredible breakthrough, I suspect that it > would simply go unrecognized and fall into the dustbin of history. To > illustrate where I am coming from... > > Half a century ago when I was graduating from high school and dinosaurs > roamed the earth, I selected 3 grand challenges to guide my early adult > years. One of those challenges was to identify the units of communication > between neurons, from which the mathematics of their manipulation could be > induced, from which the entire process could be understood, which people > like you could use to design things like OpenCog, etc. Since then I have > pretty much nailed down a significant part of this puzzle and have > discussed much of it here on this forum, probably enough to fill in the > remaining blanks, but amazingly NO ONE has shown any interest in following > this thread. > > Picking one particular tiny illustrative detail of this - my realization > that neurons MUST communicate derivatives like dP/dt rather than straight > probabilities, to be capable of temporal learning without horrendous > workarounds. I thoroughly explained it on this forum, and no one objected > to any of it, yet it has changed nothing. > > Are my efforts that breakthrough? Is there something else out there that > is that breakthrough? We'll probably never know, and similarly if something > new comes along, it will almost certainly fall into the same dustbin. > > I suspect that something like the scanning UV fluorescence microscope > could spark a massive breakthrough, but even if they were installed in labs > today, it would still probably take a decade to learn enough to > significantly help AGI. > >> >> Hawkins and others who pursue neuroscience based approaches to AGI, seem >> to me to be taking simplified models of individual aspects of brain >> function, and amplifying them into whole would-be AGI architectures. (In >> Hawkins case, what he has is probably best considered a simplified, >> qualitative model of parts of visual and auditory cortex.) >> >> For this sort of reason, I think that *if* one wants to pursue AGI R&D >> now, a neuroscience based approach is not the way to go... >> > > Every year I repeat myself on one point, in the hope that eventually it > will "click" here. Here goes for this year, with various improvements since > last year... > > There are three areas of study and development, that absolutely must > progress forward TOGETHER, because without any one, the other two MUST fail: > 1. Mathematics of cognition - the theoretical basis, which when > eventually completed sometime in the future will become the basis for AGI. > 2. Neuroscience research, to confirm/deny what the mathematicians come up > with, and to uncover new pieces of the puzzle. > 3. Simulation to confirm mathematical models. We are now in the very > early phases and call it NN. Later on when more is known, this will morph > into something that looks more like present-day ideas as to what AGI will > look like, when we start "simulating" the necessary computations and NOT > the irrelevant biological details, which we now can't even identify. > > My point here is that some deluded people (names withheld) now waste their > careers on #3 (or #1, or #2) above, outside of an environment that also > provides the others to support them. > > Here, I am a #1 person in a forum of #3 people, talking about #2 people. > Either these are all going to come together, somewhere, and proceed > forward, or others like us will STILL be having this SAME conversation a > century and more from now. > > BTW, does this mean that I need to recruit a young genius here on this > forum, to continue repeating this mantra when I am no longer able to do so, > and pass this responsibility on to some other not-yet-born young genius > another half-century in the future? Perhaps this is the fate of AGI? I sure > hope not, but present developments and participants sure don't offer much > hope. > > Hence, while I completely agree with your statement "now, a neuroscience > based approach is not the way to go", I strongly disagree with its > implication that ANY other single-pronged approach, like your present > concept of AGI R&D, is a potentially viable way to go. > > It would only take a few well-connected people, perhaps including YOU, to > realize this, get together, and create a major R&D center somewhere in the > world to pull these three disciplines (and some side disciplines like > microscopy) all together. I'd bet the Chinese might go for this. I'd also > bet that the neuroscience people would just LOVE for someone to tell them > what to look for (which has always been their biggest single challenge to > continued funding), and the mathematicians (like me) would just LOVE for > their theories to be quickly tested, *before* their papers are submitted > for publication. I suspect that you could pull all the best people from > around the world to such a center. > > ... or, you could continue pushing in your present directions as things > gradually grind to a complete standstill, fatally paralyzed by the absence > of a some key piece of mathematics, or any way to have someone else solve > your underlying math problems quickly, before your project dies of > stagnation. > > You must be familiar with the issues of combinatorial heuristics, where > workarounds start interacting with each other, necessitating still more > workarounds, whose operation eventually surpasses human understanding, > until the whole mess sets up like so much immutable concrete and can go no > further. I see NO possible way for the current concept of AGI R&D, in the > absence of a SOLID mathematical base, to possibly avoid this particular > fate. Do you? Does anyone else here? > > Wake up - and I'll see you in China!!! > > Steve > =================== > >> On Tue, Aug 28, 2012 at 5:34 PM, Arets Paeglis <[email protected]>wrote: >> >>> We do NOT know what mathematics are involved in the computations. >>> >>> Wow. Hodgkin, Huxley, Nagumo, Izhikevich and the others must, then, be >>> totally delusional. The more I know. >>> >>> [..] but we really don't know how they work and which of the observed >>>> features are actually important. >>> >>> Sure, there are still a lot of white spots in our understanding of how >>> brain works on its different levels. On the other hand, there exists a >>> massive corpus of academical neuroscience, both empirical and >>> computational, that *clearly *disagrees with the picture you apparently >>> are painting here - one where we are nearly clueless about what and how >>> goes on in our cerebral matter. >>> >>> /NJ/ >>> >>> >>> On Wed, Aug 29, 2012 at 12:11 AM, Steve Richfield < >>> [email protected]> wrote: >>> >>>> Arets, >>>> >>>> On Tue, Aug 28, 2012 at 10:00 AM, Arets Paeglis <[email protected]>wrote: >>>> >>>>> Are you seriously going to suggest that we still have no idea as to >>>>> what "neurons are doing"? >>>>> >>>> >>>> Yes. >>>> >>>> Most synapses are NOT the simplistic fractional transfer mechanisms >>>> used in NNs, We have NO idea how neurons learn as fast as they do. We have >>>> NO idea what guides their self-organization. We do NOT know what >>>> mathematics are involved in the computations. We have observed some >>>> interesting things, like some neurons becoming active under particular >>>> circumstances, but that is about all we now "know". >>>> >>>> The equivalent in astronomy: We still think the earth is at the center >>>> of the universe, that planets orbit in bizarre ways, and that celestial >>>> spheres make it all work as it does. >>>> >>>> Steve >>>> ================= >>>> >>>>> On Tue, Aug 28, 2012 at 7:52 PM, Steve Richfield < >>>>> [email protected]> wrote: >>>>> >>>>>> Alan, >>>>>> >>>>>> Your discussion fits right in with some of my postings. I have >>>>>> discussed the equivalent unity gain frequency of neurons (for 741s it is >>>>>> ~1MHz), negative feedback in the form of variable driven impedance, etc. >>>>>> It >>>>>> appears that internally, neurons may "compute" about as fast as vacuum >>>>>> tubes, and NOT at the much slower pulse rates seen at the outputs of >>>>>> spiking neurons. >>>>>> >>>>>> However, I fear that we are throwing pearls before swine. >>>>>> >>>>>> Also, I wonder if everyone else is missing an essential point. We are >>>>>> NOT saying "neurons are SO much faster and smarter that we can never >>>>>> duplicate such function on a human scale", but rather "if we make the >>>>>> effort to understand what neurons are doing, then we will have some >>>>>> chance >>>>>> of understanding the problems they are solving, after which we can then >>>>>> engineer human scale systems without being encumbered by the neuronal >>>>>> legacy." >>>>>> >>>>>> Steve >>>>>> =================== >>>>>> On Tue, Aug 28, 2012 at 9:34 AM, Alan Grimes <[email protected] >>>>>> > wrote: >>>>>> >>>>>>> om >>>>>>> >>>>>>> The elemental unit of an analog computer, akin to an NAND gate, is >>>>>>> the >>>>>>> operational amplifier. The canonical opamp is the LM741, which was >>>>>>> introduced in 1968 and is still the standard opamp that everyone >>>>>>> uses. >>>>>>> The Japanese have an equivalent part which has pretty much identical >>>>>>> specs. >>>>>>> >>>>>>> Better parts are now available but engineers usually start out with >>>>>>> the >>>>>>> LM741 and chose a different part only if it can't meet their >>>>>>> performance >>>>>>> goals. I'm kinda fond of the Jfet input opamps myself but they can >>>>>>> be a >>>>>>> bit more fragile. >>>>>>> >>>>>>> I did some computer simulations of my father's stereo (made in 1974). >>>>>>> The power amplifiers are basically power opamps made with discreet >>>>>>> parts. >>>>>>> >>>>>>> You think of an amplifier as conveying a signal from input to >>>>>>> output. A >>>>>>> classic tube amplifier does exactly that, sometimes with 6-12 db of >>>>>>> negative feedback.The creepy thing about the simulation was that the >>>>>>> signal appeared to disappear in the middle of the circuit, so I had >>>>>>> trouble even figuring out which wire was even conveying it. >>>>>>> >>>>>>> Logically, the signal MUST pass from the collector of Q402 to the >>>>>>> base >>>>>>> of Q410. However the voltage swing at that point is many decimal >>>>>>> places >>>>>>> below the DC voltage at that point. >>>>>>> >>>>>>> But that's the thing. The amplifier doesn't amplify signal! The >>>>>>> difference between the voltages at the input parts, (Q402, non >>>>>>> inverting, Q404 inverting). is in the microvolts range. (Q406 is >>>>>>> basically a 2mA CCS with a 20-30 second time delay), so the input >>>>>>> signal >>>>>>> to the amplifier is essentially null. >>>>>>> >>>>>>> But the circuit does work. It has a gain defined by R410 and R414. >>>>>>> >>>>>>> It is an error amplifier. The brain works in much the same way. >>>>>>> Neurons >>>>>>> don't say much to each other unless there is an error signal. The >>>>>>> amplifier's output is your imagination and the input signal are your >>>>>>> sense organs. >>>>>>> >>>>>>> Anyway, just a different perspective. =P >>>>>>> >>>>>>> -- >>>>>>> E T F >>>>>>> N H E >>>>>>> D E D >>>>>>> >>>>>>> Powers are not rights. >>>>>>> >>>>>>> >>>>>>> >>>>>>> >>>>>>> ------------------------------------------- >>>>>>> AGI >>>>>>> Archives: https://www.listbox.com/member/archive/303/=now >>>>>>> RSS Feed: >>>>>>> https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac >>>>>>> >>>>>>> Modify Your Subscription: https://www.listbox.com/member/?&; >>>>>>> Powered by Listbox: http://www.listbox.com >>>>>>> >>>>>> >>>>>> >>>>>> >>>>>> -- >>>>>> Full employment can be had with the stoke of a pen. Simply institute >>>>>> a six hour workday. That will easily create enough new jobs to bring back >>>>>> full employment. >>>>>> >>>>>> >>>>>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>>>>> <https://www.listbox.com/member/archive/rss/303/20912103-eed2d0e1> | >>>>>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>>>>> <http://www.listbox.com/> >>>>>> >>>>> >>>>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>>>> <https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac> | >>>>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>>>> <http://www.listbox.com/> >>>>> >>>> >>>> >>>> >>>> -- >>>> Full employment can be had with the stoke of a pen. Simply institute a >>>> six hour workday. That will easily create enough new jobs to bring back >>>> full employment. >>>> >>>> >>>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>>> <https://www.listbox.com/member/archive/rss/303/20912103-eed2d0e1> | >>>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>>> <http://www.listbox.com/> >>>> >>> >>> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >>> <https://www.listbox.com/member/archive/rss/303/212726-11ac2389> | >>> Modify <https://www.listbox.com/member/?&;> Your Subscription >>> <http://www.listbox.com/> >>> >> >> >> >> -- >> Ben Goertzel, PhD >> http://goertzel.org >> >> "My humanity is a constant self-overcoming" -- Friedrich Nietzsche >> >> *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> >> <https://www.listbox.com/member/archive/rss/303/10443978-6f4c28ac> | >> Modify <https://www.listbox.com/member/?&;> Your Subscription >> <http://www.listbox.com/> >> > > > > -- > Full employment can be had with the stoke of a pen. Simply institute a six > hour workday. That will easily create enough new jobs to bring back full > employment. > > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/23050605-bcb45fb4> | > Modify<https://www.listbox.com/member/?&;>Your Subscription > <http://www.listbox.com/> > ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
