The problem of creating an algorithm that will generate only new variations of patterns without relying on Input is interesting even though it is not AI. For example, the most methodical way would be to create some endless variety of incremental iterations on the features of the patterns to be generated. Interestingly enough, that could easily fail due to the infinities of infinities thing that can be found in a non-rational real number (or the Godel Theorem.) If, for instance, the variation iteration operated on one feature at a time, you could keep on going forever with very little differentiation if the features were designed literally to march on toward an infinite variety. The variations would presumably bog down as the program ran until it was stopped. (Ok no one is talking about real infinity but it is possible to write an iteration on a feature that could go on for years.) So in order to ensure that the pattern generator would create images that were clearly novel, no one would use an infinite iteration on a single feature at a time. If on the other hand you were to try a single iteration (of variation) on each feature one at a time, then if the program was able to generate novel features it could produce tediously similar images by generating uninteresting similar features and then iterating on them.
So while this problem does not seem to be insurmountable, my claim here is that I believe that there are a number of interesting obstacles to the problem and they seem to tie in with things like the infinite pathways of an infinitely branching selection problem. Even though I am not talking about an AI problem it seems as if it is somehow related to AI. To explain this a little more carefully. I want to create a novel visual pattern generator that will create truly novel patterns. It is perfectly ok if it uses iterative methods of variation (including incremental variation) but I need a way to test the program to make sure that it is actually capable of generating a massive number of different patterns (whatever a pattern is.) Then the problem includes devising a sampling method. But the sampling method has to be dependent on jumping around the pattern generator since we cannot wait for years while the program tediously jumps through various incremental iterations of variations. So while I want to demonstrate that the program can generate (and would theoretically eventually generate) a massive number of similar patterns (showing that it is capable of developing a diverse number of similar styles) I also want it to jump around so that it would be extremely difficult to claim that it is just doing the same kind of thing over and over again. And since I am not really sure what a pattern is, then I have the problem that it might create a large number of splotches of color that are all different and yet look like they are tediously taken from some underlying random style. So simple iteration and simple randomness will not work. Without judgement that can be learned from other thinking entities it might be very difficult to design a program that can generate an immense number of novel patterns and avoid falling into a "style" for too long a time. So the method of developing some simple visual analytical methods, (which like the pattern generator would be able to generate new features) that could be used in detection, might help me to give it some kind of innate judgement even though it could not learn anything from the Input world. Oh yeah. The non-rational real number thing: Where did that come from. Since an iterative pattern generator could fall into a sub-loop if it started repeating, it would have to be based on a non-repeating iterative method. While this does not have to be done directly, it could be done by relying on a varying number of iterators, it would have to be governed by a non-repeating method of some kind. Ergo: a non-rational real. It is a little more theoretical than most people recognize. At least, that is my opinion. Jim Bromer ------------------------------------------- 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
