Vladimir,
Yes, if a concept is defined by its associations, and if a significant subset of them somewhat distinguish a concept, it would seem only natural that links between associations of nodes A and node could help the two concepts find each other in a large, high dimensional space. This is somewhat akin to, although slightly different than, performing radial marker searches out from 2 or more concepts in a semantic space, something that dates back at least to Quillian. In a draft version of my Sat 10/20/2007 2:03 PM post to you there was a point FIVE that discussed this as a possible argument for decreasing the number of M in your prior post. But I had, and still have, a lot of very interesting reading to do, so I didnt want to take the time necessary to make the paragraph both well reasoned and reasonably well written. Beside I thought having four points against the notion that the information in the human brain contained only 10^9 bits was bombastic enough. Ed Porter -----Original Message----- From: Vladimir Nesov [mailto:[EMAIL PROTECTED] Sent: Sunday, October 21, 2007 11:34 AM To: [email protected] Subject: Re: [agi] Human memory and number of synapses.. P.S. Benjamin, It's interesting that you mentioned this right now. My discussion with Edward in parallel thread effectively led to this issue. Basically, it's useful to be able to find regularities between arbitrary pair of concepts (say, A and B) that system supports (as kind of domain-independence). But when such weak connection between these concepts is established, it's useful to be able to find more regularities between other concepts that are connected to them (between concepts associated with A and concepts associated with B). For this purpose, it's useful to impose kind of locality field on concepts that get connected. On 10/21/07, Benjamin Goertzel <[EMAIL PROTECTED]> wrote: On 10/21/07, Edward W. Porter < [EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]> > wrote: Ben, Good Post I my mind the ability to map each of N things into a model of a space is a very valuable thing. It lets us represent all of the N^2 spatial relationships between those N things based on just N mappings. This is something we all know, but it is one of the many wonderful efficiencies of mathematics we often don't stop to appreciate. Yes, a spatial index/embedding lets you efficiently get answers to a variety of queries that are inefficient to answer based on many other indices/representations.. For instance: Given X, find all entities within radius r of X ... or, find the N items most similar to X ... Thus, even for non-spatial data, it may benefit an AGI system to project data into some N-space, in such a way that Euclidean distance mimics "conceptual similarity" between data items, so as to make this kind of query efficient to answer... We have prototyped this trick in Novamente for a couple purposes... and eventually it will be integrated into the core system as a default service to be utilized by all MindAgents as appropriate... -- Ben G _____ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/? <http://v2.listbox.com/member/?&;> & -- Vladimir Nesov mailto:[EMAIL PROTECTED] _____ This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/? <http://v2.listbox.com/member/?&; > & ----- This list is sponsored by AGIRI: http://www.agiri.org/email To unsubscribe or change your options, please go to: http://v2.listbox.com/member/?member_id=8660244&id_secret=56072816-8f0314
