Well your program has some meta awareness but you can add more incrementally. For example you can write some kind of monitoring sub program and then try to figure out how the data that it produces might be more intelligently parsed and abstracted.
Jim Bromer On Mon, Feb 2, 2015 at 12:48 AM, Piaget Modeler via AGI <[email protected]> wrote: > If we're lucky, PAM.P2 will make a baby step forward. > > (Pun intended.) > > ~PM > > ------------------------------ > Date: Sun, 1 Feb 2015 22:48:06 -0500 > > Subject: Re: [agi] Multiple Conceptual Level Networks > From: [email protected] > To: [email protected] > > An attempt to consciously use coherence (more specifically than just using > some unqualified reference to weighted reasoning) is a step forward. But, I > believe the problem of replacing old information with better information is > only one case where interactions between the levels of statements and > actions on statements are needed. And I believe that even that process > cannot take place without more direct knowledge of how such interactions > can take place. So a reaction to some statements should not only instigate > highly controlled action on some statements but a meta awareness of the > action must be made available to the program and the program must be able > to 'play' around with the processes. That is something that I am pretty > sure is going to be missing in your program. > > An equivalence case can be made that under the right circumstances a > program which has constraints or severe limitations on the play between > statement and actions on the statements could still attain a higher level > of intelligence. However, (imo) the reason that this is unlikely is because > there will be such great gaps between knowledge statements. Coherence is > one way to start to deal with the problem, but it is not, in itself, going > to be enough. Let's say that it was possible to have every logical > statement that was needed to fully detail the understanding of some subject > (along with the principals of learning). Then a tightly controlled > interaction between the conceptual levels of statements and actions might > be sufficient. This should illuminate one of the problems that I have > been trying to get people to think about. With only a sketchy collection of > those statements you are going to need a lot of juice to jump the gaps. > This is where 'play' (for example) could serve as a fundamental and an > inspiration for the program. And meta awareness is a necessity. > > Jim Bromer > > On Sun, Feb 1, 2015 at 7:29 PM, Piaget Modeler via AGI <[email protected]> > wrote: > > Coherence is one way for old information to be rejected as well. Belief > parameters for coherence in PAM.P2 includes a set of accepted and rejected > propositions as well as a positive or negative coherence score. As new > information is received, it may cause old information to become less > coherent > and rejected. A cohence propagator continually revises and propagates the > coherence score for activated propositions. The coherence propagation > agent > in PAM.P2 is based on Paul Thagard's work at the University of Waterloo. > > ~PM > > ------------------------------ > From: [email protected] > To: [email protected] > Subject: RE: [agi] Multiple Conceptual Level Networks > Date: Sun, 1 Feb 2015 12:42:06 -0800 > > > So I think what you are referring to is new information overriding old > information. > > This could happen for many reasons but the essence is that some belief > parameters > (e.g., certainty, arousal, valence, activation, etc.) on the new > information cause the > new information to be considered before or superior to the older > information. > > In PAM.P2 the basic connection between statements and action is that an > intention > is formed, representing the desirability of bringing about situation (or > proposition), > and an action selector (or solver) works to find (or create) a solution > that can achieve > the situation (or proposition). > > ~PM > > ------------------------------ > Date: Sun, 1 Feb 2015 10:28:18 -0500 > Subject: Re: [agi] Multiple Conceptual Level Networks > From: [email protected] > To: [email protected]; [email protected] > > So, as I can best understand what you are saying is that: > "neural propositions as its knowledge representation" > "Propositions are disconnected from the underlying agents that refer to > and > create them. " > "There are intention prototypes and solution prototypes. Solutions are > tried > by the system, the actions of which are attempted." > > So while your program presumably has many conceptual layers and can > create more, there are constraints on the kinds of interactions that can > occur between some of the most essential 'conceptual layers' (as I called > them.) (Perhaps I should not call them conceptual layers in the context of > agi programs. Maybe I should just call them layers or something like that.) > > Suppose someone convinces a young, somewhat naïve adult that he should > forget everything he has ever learned. Of course he can't do that. However, > he would be able to start to ignore certain principals which he feels that > were taught to him but which he never fully accepted. As he goes on he can > teach himself to ignore more and more of those principles. For example, as > I recognize that an idea that I thought was my own was actually instilled > by advertising I can choose to selectively ignore it. It is my feeling that > this kind of example shows that interactions between ideas (or > propositions) *and their application to thoughts* are essential to > intelligence. I suspect that General Intelligence is impossible if an idea > about shaping one's own thinking cannot be applied. The direction for this > shaping process may require some kind of justification but that > justification will sometimes require a great deal of thought. It > can't only come from some external source of verification. > > There has to be some kind of constraints on the interactions between these > levels. The program cannot forget everything it knows just because someone > makes an imperative statement to that effect. So there has to be some kind > of buffer between the propositional level and the action level. And the > argument can be made, especially for a logical system, that the agents that > act on the propositional levels are effectively capable of doing the kind > of thing that I am talking about. (Or if they are not they can be tweaked > so that they are.) However, my point here is that the management of > something like that will introduce new kinds of problems (and situations) > that require new kinds of sub programs to work on it. > > I am going to read the paper and watch the video that you referenced. > > > Jim Bromer > > On Sat, Jan 31, 2015 at 1:27 PM, Piaget Modeler via AGI <[email protected]> > wrote: > > > > > Date: Sat, 31 Jan 2015 10:45:27 -0500 > > Subject: Re: [agi] Multiple Conceptual Level Networks > > From: [email protected] > > To: [email protected] > > > > I am reading your (Piaget Modeler's) paper, "The Neural Proposition: > > Structures for Cognitive Systems," but I am trying to reread it more > > carefully to better understand it. > > > > So let me ask you a few questions about your project. > > Is it an AGI application or an AGI Platform? > > PAM.P2 is a cognitive architecture ( > https://www.academia.edu/9997454/PAM.P2 ) > that uses neural propositions as its knowledge representation. The ovals > in the > diagram represent prototypes, instances of which are referenced by the > depicted > agents. > > > You know about reification and gerunds. How does your program turn a > > statement into an action? > > There are intention prototypes and solution prototypes. Solutions are > tried > by the system, the actions of which are attempted. attempts are sent to > a device running a psyche application and results are returned as to > whether > the attempt succeeded or failed. > > > How does your program prevent a statement like, "Forget everything > > that you know" from becoming an action that causes it to forget > > everything that it knows? > > Propositions are disconnected from the underlying agents that refer to and > create them. That being said, an agent could run amok if incorrectly > programmed > and delete all the propositions of the system. So the agents have to be > carefully > programmed. > > ~PM > > > > > > > Jim Bromer > > > > > > On Thu, Jan 29, 2015 at 12:19 PM, Piaget Modeler via AGI > > <[email protected]> wrote: > > > Do you mean like "Neural Propositions: Structures for Cognitive > Systems" ? > > > > > > ~PM > > > > > >> Date: Thu, 29 Jan 2015 06:04:02 -0500 > > >> Subject: [agi] Multiple Conceptual Level Networks > > >> From: [email protected] > > >> To: [email protected] > > > > > >> > > >> I came up with a great concept-theory using cross generalizations on > > >> logic so I decided to write about it. As I thought about it I > > >> remembered seeing some introductory text about network theory > > >> somewhere and the first examples that they mentioned used binary > > >> nodes. Some of the examples were effectively about kinds of logical > > >> cross-generalizations. So what happened to my great new theory? > > >> Somehow it fizzled into something that was from some introductory > > >> text about networks. The thing is, I don't think current network > > >> theory is very interesting. > > >> > > >> In order to create more interesting networks you have to have multiple > > >> layers. Not just multiple processing layers but multiple conceptual > > >> layers. But these concept layers should not be associated only by a > > >> simplistic associations (on concept nodes for instance) but by the > > >> potential for nodes on one layer to interact dramatically with other > > >> layers. Of course this can be implemented using contemporary > > >> conventions about nodal networks. So why is the idea of multiple > > >> concept layers important? Because of the potential of the layered > > >> networks to represent cross-categorical relations which might be > > >> needed to solve difficult problems and which might be more susceptible > > >> to effective methods of analysis. > > >> > > >> When Internet traffic is being analyzed, for example, the analysis > > >> occurs on a different conceptual level than the traffic itself. In > > >> this case, there is very limited interaction with the traffic and the > > >> analysis. If the analysis is sent to a web manager then the analytical > > >> function is itself producing some traffic on the same system. The > > >> number of conceptual levels in this example is extremely constricted > > >> (there are 2 levels) and the interaction between the levels is tightly > > >> constrained as well. > > >> > > >> But it is easy to imagine systems where there are many different kinds > > >> of conceptual levels and a lot of different ways interaction can > > >> occur. Can you do this with conventional notions about sub-networks? > > >> Ok, but there are times when you need to free your mind from > > >> conventional thinking. > > >> Jim Bromer > > >> > > >> > > >> ------------------------------------------- > > >> AGI > > >> Archives: https://www.listbox.com/member/archive/303/=now > > >> RSS Feed: > https://www.listbox.com/member/archive/rss/303/19999924-4a978ccc > > >> Modify Your Subscription: https://www.listbox.com/member/?&; > > >> Powered by Listbox: http://www.listbox.com > > > AGI | Archives | Modify Your Subscription > > > > > > ------------------------------------------- > > AGI > > Archives: https://www.listbox.com/member/archive/303/=now > > RSS Feed: > https://www.listbox.com/member/archive/rss/303/19999924-4a978ccc > > 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/303/24379807-653794b5> | > 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/303/19999924-4a978ccc> | > 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/303/24379807-653794b5> | > 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/303/19999924-4a978ccc> | > 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/303/24379807-653794b5> | > 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
