I won't be able to define the details of the logic of structural insight until I actually deal with them. At first I will choose the logic that I will use ad hoc but after going through this often enough I will be able to generalize these methods. In one sense however the logic of the method is mundane because the deterministic qualities of the methods will be found in the application. Form will follow function, not the other way around. When I find that certain methods were not general enough I expect that I will be able to make some substantial improvements in the methods as I go.
The examples I can imagine at this point are going to be somewhat pretentious or elementary because I have to rely on my imagination without actual cases from actual experimentation to use. However, the first step of learning will be derived from accumulating incidental relations and then trying to discover which relations are significant to the meaning of some subgroup of the relations and which are meaningfully incidental and so on. Using categorical substitution the program will be able to generate simple variations on the sentences and sequences of subjects and so forth. Some of these generated sentences will have meaning but at first most won't. The program will also be able to make other inferences from the surface exchanges. When the program makes an 'imaginative' connection that can represent a valid synthesis (at first these will be very simple syntheses) I will try to draw its 'attention' to the statement. The program will be able to use varying levels of complexity in its output (and in its analysis of input) and one of the methods that I can exploit is my awareness that certain kinds of exchanges can narrow in on some particular synthesis. By drawing its attention to a synthesis that I think is insightful (a relation that could be the basis for an insight) I should be able to get the program to explore that particular synthesis in more detail. At first this process will be mostly random but as the program learns it will start using previous learning as a basis for continued exploration of a new synthesis of concepts. Jim Bromer On Sat, Apr 19, 2014 at 11:38 AM, Piaget Modeler <[email protected]> wrote: > Step #2. Define your structure in terms of logical relations. > > ~PM > > ------------------------------ > Date: Sat, 19 Apr 2014 10:34:44 -0400 > Subject: Re: [agi] Structural Knoweldge > > From: [email protected] > To: [email protected] > > The traditional model of computation (the essence of the traditional > models) is that you program the formulas and then input the values. This > view can be held as sound because the computational models include formulas > to compute given values. (The input-output triplets, for example, do not > have to be added as precompiled tables. Simple computation is truly > formulaic.) > > With mathematics and programming you can begin tinkering around with the > computational methods so you might see those kinds of features added as > implementations through meta-formulas. In narrow AI (for instance) these > meta-formulas may be defined as something that is just new > programmer-defined formulas. > > With AGI you need to deal with referential values which can themselves be > used as formula-like systems (and meta-formulas and so on.) To truly embark > on a referential model your programming is going to have to be able to deal > with unexpected situations and your program is going to have to be able to > create 'programs' of some kind. I plan to use something that is > constrained to be "safe" so the program won't crash. But, the thing is, > that you cannot predefine how these programs and meta-formulas will behave. > So I plan to build parts that the program can combine. When I realize that > these parts or scripts are too limiting I will add more. But the thing is > that the program has to be able to build these conceptual scripts. > > So in a way, many formulas and program methods are well defined and if you > consider these as semi-structural then there is a great deal of information > about how structural methods work. (I realize that the term structure is a > little different when referring to structuralism in philosophy.) But > the problem in AGI programming is getting computers to learn how to build > these programs from simple components and to implement them well. > > My use of the term structure is to emphasize three things. 1. Concepts > play different roles when interacting with other concepts. 2. Knowledge is > acquired incrementally. 3. But because ideas can play different kinds of > roles there are moments when leaps of insight can take place because a new > insight can be fitted into preexisting knowledge in a way so that it holds > all together and explains how that knowledge can be implemented in ways > that the program did not 'understand' before. > > Jim Bromer > > > On Thu, Apr 17, 2014 at 2:02 PM, Mike Archbold <[email protected]> wrote: > > This is all fine, but what science is there of structure as structure? > I've been trying to sort this out recently. There are various > versions of structuralism; I think one aligned more to science and one > aligned more to the humanities. Gestalt psychology. System dynamics > / complex systems comes to mind. What else? > > It's one thing to say "the structure is such and such, and I have > these relations which are invariant." But, it is another thing to be > able to perform computations on the model which would approach general > intelligence. > > On 4/17/14, Piaget Modeler <[email protected]> wrote: > > Great minds think alike. > > I agree, in fact I have three categories: > > 1. Structural, 2. Structural Content, and3. Content. > > Once you've identified your structural relations, if you're going to > > properly bootstrap this baby, then you next need to solve the Semantic > > Kernel problem: i.e., what content relations are the core relations to > > include. > > ~PM > > Date: Thu, 17 Apr 2014 09:31:38 -0400 > > Subject: [agi] Structural Knoweldge > > From: [email protected] > > To: [email protected] > > > > There is a lot of evidence that humans, like other animals, learn > > incrementally. However, my belief is that because we use ideas in > different > > ways a new idea can interact with other ideas. There are moments when > > something that is learned incrementally can be leveraged to produce > leaps of > > insight. I call this knowledge structural because it means that an idea > can > > suddenly provide some greater structure to knowledge related to a > particular > > subject. The new increment of knowledge that triggers the structural > insight > > may or may not be the key that provides the leverage of the structure. It > > may be that some new piece of knowledge just helps to crystalize some > > structure in a way that helps the learner to better utilize other > > knowledge. > > > > In programming and computational mathematics we find distinctions between > > things like operators and operands and you have to be able to find > > distinctions between other different parts of a computation if you want > to > > use mathematics creatively. However, I think it is obvious that the > > situation is more dynamic and more fluid in thought. Some information may > > play some role based on some other information so that it can react with > > some other information and we just cannot categorize how some piece of > > information might be used before hand. An AGI program has to be able to > > find how information can work together to create greater structures of > > knowledge. But for this to happen, the program has to be designed to > provide > > the structure that will ensure that the potential to build learned > > structures is there. > > Jim Bromer > > > > > > > > > > > > > > > > AGI | Archives > > > > | Modify > > Your Subscription > > > > > > > > > > > > > > > > > > > > ------------------------------------------- > > AGI > > Archives: https://www.listbox.com/member/archive/303/=now > > RSS Feed: > https://www.listbox.com/member/archive/rss/303/11943661-d9279dae > > Modify Your Subscription: > > https://www.listbox.com/member/?&; > > Powered by Listbox: http://www.listbox.com > > > > > ------------------------------------------- > AGI > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/24379807-f5817f28 > Modify Your Subscription: https://www.listbox.com/member/?&; > > Powered by Listbox: http://www.listbox.com > > > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/19999924-4a978ccc> | > Modify <https://www.listbox.com/member/?&;> Your Subscription > <http://www.listbox.com> > *AGI* | Archives <https://www.listbox.com/member/archive/303/=now> > <https://www.listbox.com/member/archive/rss/303/24379807-f5817f28> | > 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-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
