*So – straight question – what are you saying? – we need logic plus adaptations for complexity? * ** Well, I already said that AI needs IO and that logic is inadequate. I feel that the substantial argument for GA's and the argument for NN's are that they can create novel combinations (of referents to the IO Data Environment) which can then be used and tested for effectiveness. I do not feel that GA's or NN's are - in themselves - adequate for genuine AGI either. There are many ways we could create novel combinations of components and test them for effectiveness.
When I talk about combinations of components I am not just talking about pre-defined combinations. So, looking at the visual field, one can create novel combinations of pixels to represent something that you have never seen before (ie a new image that captures your imagination partly because it is a little unusual in some way.) So while the color values of a pixel may be predefined it should be clear that there are combinations of pixels that would create an image that you had never seen before. It is true that microscopic colors (like oil paint) can create combinations that go far beyond anything that a pixel can represent, but if we wanted our pixels to represent those microscopic particles of paint we could combine the concept of the pixel with the concept of the relations of scale and, given enough data space, create a microscopic approximation to the painting. (I once got in trouble at the Metropolitan Museum of Art in NYC because I wanted to get my eyes as close as I could to a Vermeer. -There are only 17 known Vermeer's in existence.- I wanted to see if the image became more cartoonish as I got closer and I found that it did. But I had to get my eye right next to the painting before it became obviously cartoonish. I wasn't able to see this in the images of his paintings that I had access to at the time.) So scale is something that can be important in some cases. So, the microscopic particles of paint (and oil or other base) will hold far more information than one set of pixels, but by combining and appropriately relating the cogitative representation of 'scale' with the IO representation of 'pixel' we can approximate that level of detail. The simple fact is that if you casually look at a Vermeer you won't be seeing the microscopic mixtures of particles anyway so rough approximations should be enough as long as you have the opportunity to make better approximations as you need to. So the only thing I would say is absolutely necessary is a facility to create novel combinations of components, a method that can create the relations between combinations to make it so they can be used effectively for some purpose and the ability to create novel components from this process. I believe a substantial improvement in combinatorial complexity would make more powerful AGI feasible. However, a substantial improvement would only require a small step in mathematics. It is a difficult problem to solve but it is a simple problem. I don't think you, Mike, have seen how this could make a difference but it would just because it is such a primal problem. If it is a solvable problem then it is a problem that permeates all areas of computer science. But, the thing is, the application of a solution would show an exponential increase in magnitude for big data problems as compared to the way things are done today. And AGI is a big data problem. Jim Bromer On Sat, Nov 10, 2012 at 8:49 AM, Mike Tintner <[email protected]>wrote: > *Jim: It seems obvious that pure logic is insufficient to represent AGI* > ** > *So – straight question – what are you saying? – we need logic plus > adaptations for complexity? * > ** > *And, if anyone else is interested, what are the other basic positions on > this? We need the cognitive synergy of yoking separate algorithms, maybe > incl. GA’s (Ben)? What’s Richard saying (I’ve forgotten) ... we need > stochastic processes too ?* > ** > *Everyone wants to take current technology and extend it with some other > stuff – fair?* > ** > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/10561250-164650b2> | > 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-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
