Tony Lofthouse wrote: > Is there any value in having nodes located in a virtual n-dimensional > space? > > One of the advantages that artificial systems have over biological > systems is that the cost of long distance links/synapses is no more > expensive that short distance connections. (Where cost is directly > related to distance between nodes/neurons.) Is this a good thing though? > > > Would a way to calculate cost of connection using distance be useful?
Not surprisingly, this is an issue we discussed at length a long time ago (in 1997 when architecting the very early versions of Webmind). I am pretty sure that imposing a 3-dimensional spatial structure *generically* on Novamente nodes and links [or on neurons in an NN AGI architecture, for that matter], is not a good idea. In most aspects of intelligence, this structure imposes a lot of arbitrary restrictions, which then need to be worked around in convoluted ways. However, when one is looking at specifically representing the external 3D world and actions in it, there's obviously a stronger argument for using this kind of structure. The question is, what's wrong with just going ** PixelInstanceNode: P ConceptNode: R EvaluationLink (PhysicalLocation R) ( P , (4.3,5) ) *** where R is a label for a certain subjectively-considered 3D-space, and (PhysicalLocation R) is a location predicate relative to R. The "3D embedding space" you're suggesting is basically a way of indexing these EvaluationLinks by the coordinates (e.g. (4.3,5)) for rapid lookup [and consequent rapid manipulation]. I think that this kind of specialized indexing may indeed be useful for dealing with PixelInstanceNodes, but not for nearly all other kinds of nodes & links. -- Ben ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/
