I think that the simple differentiation of programs as narrow AI or AGI is overly simplistic. If there has never been an AGI program then that would indicate that there probably never will be, or at least not with contemporary computers. Instead we need to be able to define sub-programs as being Narrow AI or AGI. For example, an AGI subprogram might call Narrow AI functions (or subprograms).
Can we distinguish AGI subprograms from narrow AI and non-AI subprograms? I think that has to be done if you are going to attempt to define AGI in a practical way. I am going to refer to a subprogram as a function in this message even though it might be a little confusing. A function that does nothing other than output a value is not a Narrow AI program. A Narrow AI function has to produce outputs that are dependent on learning and which can potentially be used in further learning. This means that a Narrow AI function must be closely related to some other system of learning in the program. Therefore, a function that might be Narrow AI in one program might not be in another program unless the resultants could be shared with a different program that could use them in learning. In order for a Narrow AI function to be used by an AGI function it would have to be used in some way which would require and produce some greater powers of judgement. This means that the definition of an AGI function, while dependent on a potential of further AGI actions, cannot be defined simply by that dependence. So an AGI function has to be produced by judgement-guided learning and it has to be potentially useful in further judgement-guided learning. I have a preliminary definition of artificial judgement so this definition works for me. However, many people have disagreed with my definition. I define judgement as a process of decision or contemplation which is dependent on many learned processes. This definition is not quite strong enough because a simplistic logical decision process could be qualified as judgement. I came up with the Conceptual Typing theory to distinguish between simple logical or other simple mathematical functions and AGI learning. So a requirement of AGI judgement is that a potential variety of kinds of Conceptual Types have to be used to make a decision. To give you a simple example a causative relation is not defined by logic alone. (You could define a logical process by reference to a relation of causation but you need that reference and a potential to discover other relations dependent on it.) A Conceptual Typing not only allows for a programmer defined typing of a Concept but it also allows for the dynamic definition of a Concept Type as well. So judgement has to be dependent on the application of Conceptual Types. This is not quite strong enough but it is a start. I hope this helps someone, other than just me. -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
