Todor, On Wed, Apr 3, 2013 at 1:12 AM, Todor Arnaudov <[email protected]> wrote:
> One thought from me: throw away NL for representations in an intelligent > system, > Of course. I guess I didn't make it clear, but I was looking for an internal information exchange format to turn NL into, that would be useful. Clearly, any form of representation (like NL) that is nearly beyond the capabilities of modern computers even to parse is completely worthless as an internal representation. I worked on the supercomputer compiler at CDC, that had a related problem. They were taking FORTRAN and teasing out the array operations that the DO loops were doing, and then having their array processor do those same operations. First they parsed to parse-trees. Then they generated pseudo-code from the parse trees. Then they looked for loops that met various complex criteria indicating that they were suitable for vectorization, and translated that pseudo-code to vector operations. Then they generated sequences of machine code to replace the pseudo-code. Then they optimized the extraneous instructions out of the program. At each juncture, the program was described according to various structures. Each pass moved the program to the next structure. The final "structure" was executable code. I suspect there is a similar process that would work for NL, but before we can define that process, we need to know what is supposed to come out the other end - the equivalent to the "executable code" that a compiler turns a "program" into. AGI has a special problem in addition to the many problems shared by other applications like automated language translation. AGI needs the ability to COMPUTE on the results of "compiling" the NL, which is even more closely parallel to a conventional compiler than other NL products. So, what the heck can we compile NL into that would support prospective AGI operation? If this question can't be answered, then it is time to fold up the AGI tent, shut down this forum, and give up. Right? This whole thing SO reminds me of my first robotics project. I was just 5 years old. I nailed some boards together in the shape of a body, legs, arms, head, etc., and proclaimed that I was done, with the exception of its musculature and nervous systems. I STILL haven't been able to finish off those "little" details. It sounds like OpenCog and other efforts are doing the same thing - hammering the easy stuff together and expecting to take care of the rest later. My wooden robot had NO suitable interfaces to attach future musculature and nervous systems, and neither does OpenCog and other efforts. Clearly, if no suitable internal interfaces can be created, then the code to honor those impossible interfaces can't be written, so there is absolutely NO hope of the easy stuff ever amounting to anything. Right? Ben, are you following this? Steve ================= unless there's a mind that does understand it completely - e.g. one that can unroll and map any ambiguous vague NL item (as long as it falls in its cognitive reach) down to sensory data specifics or/and theory/implementation details; and back. > > > >Date: Tue, 2 Apr 2013 10:07:18 -0700 > > Subject: [agi] What is "understanding"? > > From: [email protected] > > To: [email protected] > > > Aaron, et al, > > > Recent discussions regarding representation brings up an even more > fundamental question - what is "understanding". > >(...) > > >I am NOT looking for vague wishy-washy words. I am looking for a solid > definition that defines the outer boundary of > >"understanding", enough to guide efforts to define a good representation > for it. > >Any thoughts? > > > -- > ===* Todor "Tosh" Arnaudov ===* > * > .... Twenkid Research:* http://research.twenkid.com > > .... *Author of the world first University courses in AGI (2010, 2011)*: > http://artificial-mind.blogspot.com/2010/04/universal-artificial-intelligence.html > > *.... Todor Arnaudov's Researches Blog**: * > http://artificial-mind.blogspot.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> > ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
