In principle, a neural net can be built to do anything -- neural nets are Universal Turing Machines
However, the human brain has 100 billion or so neurons, and a quite complex architecture with multiple parts each having its own structures and chemical properties, and its own billions of neurons, and connected together subtly It's hardly surprising we haven't managed to replicate many of its functions using neural nets that are orders of magnitude smaller, and with very different connectivity patterns (since we don't know the brain's connectivity pattern)... IMO, neurons are just the element brains happen to be built from. There's nothing special about building systems out of neurons, as opposed to some other reasonably flexible dynamical "computing element". The magic ain't in the neurons, it's in the connection patterns, the learning heuristics, and the ensuing complex emergent dynamics. -- Ben G On Tue, Jul 10, 2012 at 9:14 PM, Jarrad Hope <[email protected]> wrote: > Hi Ben, > > It seems strange to me that NN's aren't effective for dealing with reason, > logic and language, etc - after all our brains seem to be capable of it? > Perhaps current research is not taking the right approaches / our > understanding is too limited? > > Good to see it has some applications though. > > > > On Tue, Jul 10, 2012 at 12:39 PM, Ben Goertzel <[email protected]> wrote: > >> >> Look at Juergen Schmidhuber's work, as a good example of RNNs pushed to >> the limit of current understanding... >> >> Recurrent NNs, like other purely subsymbolic methods, have not proven >> convenient (or maximally effective) for carrying out tasks involving logic, >> language, reasoning, etc. ... >> >> For tasks involving sensory perception, motor control or data >> classification or nonlinear regression, NNs are powerful and demonstrably >> applicable -- but, tend to take more computing resources and (importantly) >> require more case-by-case tuning and fiddling, than other methods.... >> >> For instance, >> >> -- to do supervised data classification, I'd choose SVMs over NNs, on >> grounds of speed and auto-tunability >> >> -- to do vision processing, I'd choose a non-NN hierarchical architecture >> like DeSTIN over an NN, on similar grounds >> >> ... ben >> >> >> On Tue, Jul 10, 2012 at 5:54 PM, Jarrad Hope <[email protected]> wrote: >> >>> Hi, >>> >>> I was wondering if anyone could point me to some resources on developing >>> AGI with neural networks? >>> (variants of ann and learning algorithms) >>> >>> So far I'm aware that it should be a form of RNN - I have been looking >>> at BlueBrain alot - but I feel that modeling a mammalians brain in that >>> much detail is better suited for WBE rather than a strictly AGI goal. >>> >>> Also why aren't ANN's used almost exclusively in developing AGI - is it >>> a memory/computational power hurdle? or is there some other limiting factor >>> I am not aware of. >>> >>> Regards, >>> Jarrad >>> >>> *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/303/22581513-9fe46a1c> | >> 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 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
