Ben, On Tue, Jun 19, 2012 at 8:17 PM, Ben Goertzel <[email protected]> wrote:
> > Steve, > > I understood your point. > > However, just to be clear, my point was that > > (A) > > " > -- Capability or method X, if you could do it incredibly (i.e. > unrealistically) well, would enable arbitrarily great general intelligence > -- Simple versions of X, seem to lead to interesting "narrow AI" behaviors > " > > does not IMPLY > > (B) > > " > -- By pursuing more and more complex versions of X, we can get high > levels (e.g. human-level) of real-world general intelligence > " > > However, just because "A does not imply B", it does not follow that "A > implies not-B". > > That is,: just because A holds, this doesn't imply that B will not be > true... > True - there is no proof that your approach cannot work. > > My reasons for thinking that OpenCog will work for AGI, are not (A) ... > even though (A) happens to be true for OpenCog. My reasons for believing > (B) regarding OpenCog are different ones, based on the theory of mind > underlying Opencog. > If you can figure out every single detail, and if you live long enough to do so, then you will succeed. However, there are several imperfect quasi-proofs that this will be VERY difficult. > > Does this clarify my perspective? > I think that I have long understood it. Regarding your rejected AGI conference paper, as I recall it was about a > brain scanning design of yours. Evidently your idea was that AGI should be > approached via detailed brain emulation, Not at all, though it might someday be used for that. However, long before that day comes, the answers as to how we "compute" SO many things that are not now understood should emerge. Besides, I suspect that the knowledge to build Ben-style AGIs MUST be in hand to ever be able to debug a simulation approach. but we don't have good enough brain imaging tools yet, so we should focus > on building those imaging tools instead of on AGI directly. My point was that this is low hanging fruit - that diagramming brains is difficult, but not as "impossible" as many people presume. Mouse brains could be completely diagrammed by commercial systems that could be produced in maybe 3 years. > We did give the paper to some experts on brain scanning to review, and > they weren't compelled by the idea. Of course, experts can often be wrong, I never heard any significant technical objections, just that it was off the topic of the conference. > but that's how refereeing works, as you know... > In any case there is a whole community of researchers focused on > computational neuroscience modeling , and brain imaging, and so forth. > > So, your main points seem to be > > -- AGI researchers should be doing brain scanning and computational > neuroscience instead > More precisely, someone should be doing the scanning, and enough computational neuroscience to clarify what might be happening to connect form with function. AGI has many unanswered questions, and so they should participate in this process. > > -- But they should be doing brain scanning and computational neuroscience > your way, instead of the way those fields are doing things > The tools don't now exist to do real-world brain diagramming, so there is no way to ever start doing real-world computational neuroscience, so these fields don't now exist in any useful form. Without this support, there are probably dozens and perhaps hundreds of unknown and unsolved computational challenges before AGI can become real-world. I just want to get the long process to closing on these issues started. So long as this remains in the closet, Kurzweil's exponential curve for AGI completion probably hasn't even started. > > If these are your main points, I think you're barking up the wrong tree > here. You really should be taking those points to mailing lists on comp. > neuro or brain imaging, right? When "selling" something, it is necessary to talk to the "customers". Here, the AGI community are the prospective "customers". Perhaps after another decade or so of hitting the brick wall, even you might be ready for more real-world information as to how we work!!! After all, those folks already agree with you about what is most important > to work on, so you only need to convince them of your particular methods... No, those are the guys who will build the subsystems. Unlike AGI, there are no "missing technological links" to building a brain diagrammer - just the resolve to fund, build, and use it. Whereas with AGi folks, you have to convince us that both your focus > *and* your methods are good, which is gonna be harder. Ben, I see your argument as being VERY similar to that of a C++ programmer who claims no need for debugging capability, as he can surely either write correct code to begin with, or figure out what is wrong without debugging software. Much of the time this is possible, but it only takes one subtle memory-clobbering bug to doom such efforts. Here, you have "low level debugging" provided by the various systems, but are rejecting the need for the "high level debugging" that a sneak peek into brain diagramming could provide. Lotsa luck, but you will have to be extremely lucky for this to work. One nasty problem (from the dozens/hundreds that lie ahead) that proves to be beyond human understanding, and the entire AGI approach and all those who participate are doomed to fail, regardless of how diligently they work. If this should happen, AGI will probably follow the Perceptron path. Everyone must forget who you are/were before any efforts could then restart, and the term AGI would never be applied to such efforts, just as no one now applies the term Perceptron to anything that is now being done. It looks to me like you (and the reviewers) are needlessly betting the farm for damn little prospect of return. Lotsa luck, because you'll need it. Steve ------------------------------------------- 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
