On Sun, Dec 18, 2005 at 02:22:27PM -0500, Pei Wang wrote: > > Biological cognition is based on network processing, too. > > No problem here --- it is in the "NN ideas" that I think is necessary. > However, it doesn't only belong to neural network, in the technical > sense. Both Novamente and NARS do network processing, in the broad > sense.
I'm unfamiliar with their architecture, so I can't comment. > > Because you're reading this message in realtime NNs are clearly a quite > > powerful model. In fact, since we don't have any human-equivalent symbolic > > processing systems, the burden of proof is reversed. > > I've addressed the difference between the natural NN and the > artificial NN. Do you mean that ANN is already "human-equivalent"? Of course not. I thought I made that very clear, but obviously failed to express the sentiment properly. Current ANNs are a particular randomly picked region from the vast automata network space. They're modeling some aspects of biological tissue information processing properly, but missing many critical components necessary for an AGI. For some reason, people tend to latch onto these limitations of particular instances as intrinsic to the whole idea of automata networks, which is quite silly. > > The space of automata networks is vast and almost utterly barren. > > The probability of hitting a fertile spot with an educated guess > > is basically nil. Of course biological tissue processing does some > > fancy tricks ANNs can't yet. > > An approach of research is not necessarily more fruitful than the > others just because in it there are more unexplored possibilities. The point is that we have a number of working instances (us), a process which resulted in such (evolution), and can map the parameter space of automata networks with an effectively brute-force effort to find fertile regions, or analyze and parametrize existing living instances (hopefully, no gentle readers will be harmed in the process). > > There won't be any AGI on any current memory-starved, few-threaded > > hardware. > > Of course, the current hardware is not designed for the purpose of > AGI, and any successful theory will even lead to new hardware, which > will support more powerful systems. However, I haven't seen any > convincing argument that lead me to agree that the key to AGI is new > hardware design. Actually, all the previous attempts on that direction The arguments are very simple: 25 GBytes/s memory bandwidth, and 10^10 nodes with some 10^6 computational elements each, with a ~ms resolution range result in requirements not compatible with what we know of physics of computation. So you have to formulate your problem in terms of 10^6...10^10 asynchronous, locally-tight globally-loose coupled threads, which will require fine-grain parallelism. No current systems (especially, affordable ones) do fit the bill. > lead me to belief that without a good theory, people don't even know I intensely disagree that there is a good, simple theory you can write down on a piece of paper (or even a stack of sheets of paper) which can be understood by a single person. This isn't at all like Navier-Stokes. > what hardware is needed for intelligence, except simple ideas like > "massive parallel processing, huge amount of memory, and superfast > CPUs", which I all agree, but there are simply not enough to answer > all the questions about intelligence. Evolution is a very simple process, fundamentally. It can result in sentient readers of this document here, if given enough runtime resources (population size and enough iterations). Similiarly, a Blue Gene under every desk allows parameter spaces to be mapped, reality simulators to be run in realtime, and populations of networks controlling virtual critters be screened for problem-solving skills, and be generally let to co-evolve. All in absence and cheerful disregard to a generic theory of AGI (which I don't think even exists). -- Eugen* Leitl <a href="http://leitl.org";>leitl</a> http://leitl.org ______________________________________________________________ ICBM: 48.07100, 11.36820 http://www.ativel.com 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
signature.asc
Description: Digital signature
