It is important to try your ideas out of course. So just trying stuff out and seeing how it works is a fundamental and essential stage of experimentation. However, I am saying that you have certain central ideas which call also probably be tested in stages. And you want to be able to examine the results of your experiments as they can relate to your central goals. So even though you might be able to add features incrementally, you really want to test simplified models of your central ideas as those kinds of tests become feasible. You might not have a precise plan but you are thinking about it in the terms of expectations that you have about the program as it gets past stages of development where a number of important features have been implemented. I am suggesting that you should try to keep brief records of your expectations about the development of your project just to keep yourself honest. And if you find that you have to keep kicking your expectations down the road then one day it may be time to change something in your plans. For instance, I have to follow through and just try some of my ideas out this next year so I will be able to gather some evidence that either supports their feasibility or suggests that they are not feasible. However, I have already accepted the idea that AGI probably requires a method to make logical complexity somewhat more simple which means that I don't have it all figured out (even if it turns out that my ideas could be used with technology that will be available 50 years from now). For example, I believe that I will be able to teach my program a rudimentary human language. The evidence for this is that I will be able to use the language to teach the program certain ideas using the terms of that language. However, that does not mean that those ideas will be integrated in the way I think is necessary for actual intelligence. So this means that I believe that I should be able to teach the program a rudimentary human language that can act like instructions to suggest relations between word-data-objects. So right away I will have an early feasibility test. But, there is the possibility that if the language turns out to be like an extremely simple database language then it will not prove that I was actually able to teach it a rudimentary human language. But, because I was able to deduce that problem before hand I can now add a characteristic to the language that I can test for (or at least which I would be able to distinguish during the testing.) The language cannot just be a keyword computer language that has a unique definition of relational meaning for the usages of each contextual term. So the language will have some ambiguities but even more to the point the 'database' operations that are (learned to be) associated with the words, phrases and anaphora-like-connected parts of sentences will possess different relational and categorical implementations (defining how the word-based-objects are related in the much simplified AGi database.)
So this means that I will be able to conduct a feasibility test of one of my essential concepts before the program is more fully developed. This is important because I am not confident that the more insightful integrated AGi is feasible because it would be combinatorially complex. So, I was able to take the idea of narrow AI and recognize that it would be easier to 'teach' a computer (through incremental trial and error) to acquire narrow rudimentary data base instructions then it would to teach it a more human-like language. Then I was able to take my conjectures about conceptual roles and structures and apply it to the characteristics that I believe would distinguish between a rudimentary database language and a rudimentary human-like language and from there I can create feasibility tests that would demonstrate minimal feasibility and demonstrate minimal effectiveness feasibility. My conjectures are now dependent on my presumption that I have discovered a way to distinguish between a simplified database instruction language and a highly simplified human-like language that could be used for database instructions (to define the relations between word-based-objects). My presumption might be wrong. Maybe my idea of a highly simplified human-like language (one in which word-based-objects can play different roles) is somehow flawed. (For example, perhaps my most simple definition of a human-like language is too simple.) But the fact that I have been able to take this discussion somewhere where it has never been and the fact that I should be able to make a minimal feasibility test of these theories would add credence to the conjecture - if the tests were successful. Then I would have to continue the to develop the program to see if I (or someone) could create a more compelling test - before the program gets bogged down by combinatorial complexity. I have started to define my AGi model in ways that are somewhat novel and I have developed early feasibility tests that would show minimal feasibility and minimal effective feasibility. If someone does not understand what I am talking about I probably would not be able to convince him even if I felt that these first two tests were successful. However, there are some enthusiasts who would be able to understand what I am talking about and my experimental tests would be persuasive to some of them (if the tests were successful). Some of the enthusiasts might be able to suggest secondary effectiveness testing even given a small window between minimal feasibility of concept and minimal effective feasibility that the concept could be used to produce an AGi program. This is a description of a part of the scientific method. It is a little difficult to describe minimal effective feasibility now, but it would probably be easier when we have more to work with. So, for now I have to emphasize a theoretical construct of conceptual roles and structures (which unfortunately no one other than me really accepts), and try to explain that this can be seen as something that is different from an imagined minimal database instruction that could assign simple relations between word-concept-objects and retrieve those related objects in a way which would look like it was almost insightful. Jim Bromer ------------------------------------------- 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
