I finally found what I was looking for. Contemporary AGI programs suffer from complexity problems. The more information the program is given or has attained the more difficult it can be to find the right data for the situation. Unfortunately, this complexity problem seems to affect all stages of assessment and so existing AGI programs fail at even elementary levels of competency. This is unacceptable. I came to the conclusion that animal minds must either have a sophisticated system of solving complexity problems or that intelligence was itself the solution for most complexity. The first does not seem likely but without even elementary competency I could not find evidence for the second. I could make an argument for either case but now I think I have insight into a possible solution for elementary competency. As I wrote my summary I realized that I was constantly contradicting my own opinions. For example, I do believe that if a problem is super complicated then you need to study it carefully before you get enmeshed in it. Yet I am also critical of becoming overly preoccupied with purely abstract generalizations. We should not expect too much from an elaboration of pure conjecture. The generalizations and the programming should be derived from the experience of working with an extensive number of cases. These principles can be easily integrated, but there was something that still bothered me about this. I could not really highlight it until yesterday. Complicated problems which do not lend themselves to the discovery of solutions through trial and error have to be carefully studied. We need to bring rational creativity to those kinds of problems. Rational creativity, where possible solutions are designed according to a better knowledge of the characteristics of the problems, can enhance the likelihood that incremental trial and error methods will work. The conjectures have to be developed from the study and implementation of individual cases. But, contrary to an implication of my summary, many of us have spent a great deal of time thinking about the application of their theories on real world problems so I could not understand why I bobbled that part of my summary so badly. I now recognize that human beings and other animals have methods to deal with ‘ideas’ and ‘concepts’ of the mind and these need to be considered as well. I think that the ways we deal with ideas and concepts have to be studied more thoughtfully and good AGI methods (sub programs) need to be developed to emulate some of these abilities. So while the major AI paradigms have been applied to real world problems they have not been insightfully applied to these hidden systems of how we work with ideas. I believe that this issue may be a part of the best way to differentiate between “narrow AI” and AGI. Narrow AI programs are unable to deal with ‘ideas’ and ‘concepts’ in ways that emulate or approximate how human beings do. As I have explained there has been a great deal of bias directed toward the idea of ‘ideas’ as a valid psychological theory or as a valid computational theory. This is why there is not much validity to the criticism that the summary was just cog.sci stuff. Developing the programming to work with concepts and ideas is a major part of what will distinguish AGI from Narrow AI. The lack of recognition that there might be systems of the mind that deal with the usages of ‘ideas’ and 'concepts’ is based on an outstanding academic bias. The implications that my summary did not describe any computational implementations shows that the bias against the study of ‘ideas’ is so well rooted in our scientifically aware society that it is unconscious. And any theory that goes against the predominant bias is going to sound like a manifesto of some sort. One criticism that was made was that my summary wasn’t even a theory. This is relevant to what I am saying here. The principle that a ‘theory’ has to be based on ideas which the experts in the field can test in actual experiments is somewhat like logical positivism. You might call it experimental positivism. A theory is not a theory unless it can be tested through experiment. We also have paradigm positivism. A theory is not a theory unless it can be experimented using our paradigm. (Experimental positivism and paradigm positivism are perfectly ok but they cannot be taken as universal tests of whether a theory concerning the subject is appropriate for the study of the subject.) Well, anyway, my conclusion is that, yes I agree that I did not provide enough details about implementation in my “Summary of my Theories About AGI,” but the summary did provide some sketchy implementation considerations about important issues. So to summarize, I believe that an emulation of how human beings work with ideas and concepts has been seriously missing in AGI. While some methods were implicitly studied in the attempts to develop AGI programs, a careful examination of the conceptual development has been unconsciously and consciously suppressed during the last century. The difference between “narrow AI” and AGI can be described as an ability to work with ideas in ways that approximate the creativity and insightfulness that human beings and animals display. If my theory is right, then my highlighting of this new area of research should give me an advantage that other contemporary AGI research lacks. So the summary of my best ideas about AGI: An AGI program is going to collect a lot of data. My guess has always been that about half the database needs to be dedicated for providing indexing into and across the other data. Artificial Imagination and creativity is absolutely necessary for AGI. Typically, the best application of imagination will be done through rational creativity where selections of ‘concepts’ to be analyzed and synthesized will be governed by relying on categorical relations. The typical scenario of recognition or analysis is a back and forth process where input is analyzed to find good possible matches with previously acquired information, and then the selected previously learned information will be imaginatively projected back onto the input to see if it can be used to explain the data. Then this process will be refined and repeated until a fairly good interpretation of the data has been found. One of the most productive forms of learning is accomplished through structural integration insight, where a component of knowledge can be used to explain many different things at once. This structural insight can then be used as a part of recognition, analysis and reaction.
But the missing tier of contemporary AGI is an emulation of how human beings (and other animals) work with concepts and ideas. 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
