I would be more inclined to assign nodes to concepts, classes, and specific instances (objects or events) rather than words themselves, since often words simultaneously represent multiple distinct concepts or classes as well as their instances.
This means, however, that the system must be able to automatically identify the use of distinct meanings for the same word using context cues surrounding it's use, which can be a rather complex issue. Currently I'm focused on using fuzzy connection strengths for edges in the semantic net, along with rules for how those connection strengths interact based on local network topology. (This sort of approach has significant advantages in dealing with ambiguities and conflicting information.) For example, an instance may have multiple edges indicating which nodes represent its class. A collection of local daemons would be triggered to help resolve this scenario by strengthening or weakening the connection strengths until a clear winner or winners had been found. I still have a ways to go in building the mechanisms to convert a semantic representation of a sentence in terms of word relationships into one that describes the meanings of those words. I look forward to actually testing my ideas out. On Aug 24, 2012 6:19 AM, Jim Bromer <[email protected]> wrote: If I was writing an AGI program I would start by designing a program that developed a semantic net as well. Yes, as the program uses a concept, whether it is reacting to it or developing an idea around it, whether it is the in the focus of the idea or closer to the periphery, once it is considered or used in a consideration it may change in response to other concepts that are used with it and it can change the other concepts as well. And I feel that this is how we use concepts. As a result the well defined concept (or word in the case of a semantic net) becomes much more varied and conceptually fuzzy. If you stress too much formality in your definitions of concepts you are going to run into some confusion just because the meaning can carried along by the way it is used and by the way the it can be used. On the other hand, an AGI program has to have some way to "understand" the concept. So the only way that this can be done is by looking at different usages, and becoming aware when a usage goes crosses the intuitive boundaries that you or others have held to recognize that it could be understood in other ways as well. This sounds relativistic to me. Other aspects of relativism point to the the lack of a fundamental base from which to work on. Even though there may be fundamental truths in an environment, that does not mean that we can truly and fully understand them. So when someone starts talking about a program that uses Bayesian Reasoning, for example, without specifying the frame which it would be used in (whether that specification represents an instance or an abstraction meant to represent possible instances) then I become wary that the person is talking about an idea that would work in narrow AI but might not work in AGI where important methods have to be stronger. Of course without any fundamental bases for reality one can argue that no definition is adequate, but that is not the point. We have to do something to make rough representations of fundamentals, but these fundamentals are not invariant universal absolutes but more like specialized concepts that are useful. Jim Bromer On Thu, Aug 23, 2012 at 10:53 PM, Aaron Hosford <[email protected]> wrote: Ah, that makes it a lot clearer. So really it's about experience and context (including other concepts and your own personal usage) modifying concepts on the fly. A dictionary always defines words in terms of other words, and so as the meaning of one word shifts, so do the meanings of all those connected to it. An AI can manage this sort of difficulty if it is built on the same principles. That's why I prefer semantic nets. They represent a data structure that corresponds directly to this sort of shifting web of concepts. Each concept receives its own node, and is defined entirely in terms of its connections to other nodes, which are updated constantly as new experiences and thoughts accrue. On Thu, Aug 23, 2012 at 9:09 PM, Jim Bromer <[email protected]> wrote: Aaron Hosford asked me if I could give an example of Conceptual Relativism? No. I mean that everything is an example and conceptual relativization is so omnipresent that we are constantly adapting around it. Let's say that you want to understand what a word means. You might start by using it in some sentences. But every time you use it in a particular sentence your sense of the meaning of the word naturally becomes more strongly associated with that situation. So you try to define it in more sentences (that is you try to better understand the concept-word by using the concept in another situation.) It then takes on the characteristics of that its specialized usage as it relates to that situation. Furthermore, you realize that some language-concepts can only be defined by systems of words and the meaning of many of the words has to be fitted to the specialized system of the sentence or phrase. The same thing is true for most any concept that you want to think about. Ok here is an example: What does pattern mean? Well, maybe someone might start with saying it is a symmetric image that is repeated. Then you start to question what a pattern is and look it up in a dictionary. It is not necessarily symmetric. By including the concept of a meta pattern into your definition you realize that pattern is not necessarily graphic. Then you realize that there can be different definitions of meta patterns. You realize that there are different -types- of patterns and meta patterns. And on and on. At some point you realize that your new definitions of a pattern may have taken you beyond the boundaries of what you would intuitively call a pattern. So then you have to ask yourself if the new definitions are valid. If you start by asking if it is useful or interesting then the new variations will become more acceptable because they are interesting. If you find that there is a better term to apply to some of the new variations you can learn to accept the fact that even if they are related to the concept of a pattern it might be easier to get other people to know what you are talking about using the better term. But then you realize that in some conversations the term 'pattern' can help people relate to the greater context of some particular situation that you are trying to describe. I think this is a pretty good example although I did not present it as well as it could be. It is not just a language thing, it is a subject-of-thought thing. In fact, you can sometimes rely on social convention of language to tone down linguistic relativism, but because conceptualization might tend to include more systems of related thought then you can describe using language the social conventions may get in the way of the effort to better understand an idea. It is my opinion that our knowledge is permeated with conceptual relativization and as a result the effort to limit the use components-of-knowledge in thought becomes complicated. In fact, the components-of-thought model becomes very tangled and when we use components-of-thought they become more like indexes into a range of variations on how the concept is used with other concepts. I think the most efficient computational methods are like combinatorial component systems. I would want to use combinations of component concepts because that seems like the more efficient method for AGI. But, because of conceptual relativism this system cannot be logically constrained according to universal principles. Of course the concept of 'universal' in logic is relative. Jim Bromer On Thu, Aug 23, 2012 at 8:16 PM, Aaron Hosford <[email protected]> wrote: OK, that gives me a partial grasp. Can you give me an example? On Thu, Aug 23, 2012 at 7:05 PM, Jim Bromer <[email protected]> wrote:Conceptual relativism is the idea that concepts must be used to think about other concepts and when that happens the concepts that are used in an expression or study of the subject concept can often affect the "meaning" of the subject concept. So concepts are not only relative and relational they are also relativistic. AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
