I felt that my comments on algorithmic synthesis were useful. We can imagine an AGI program that does not understand much about a certain object (of thought) becoming aware of the possibility that it might be used in a synthesis if it was aware of a feature of the object which it thought might be useful when strongly bound or associated with another object. However, this case is like the many traditional AI examples which seem so theoretically sound but which are much more difficult to use in actual cases. For example, suppose that we knew someone was fast with an answer. Then, our AGI program might deduce that he or she should run in a foot race that we had been talking about (if it knew that we were looking for someone fast to race for our team). So it becomes obvious that this method, which is a good one, is fraught with the usual complications. So before the program assumed that the synthesis was a sound one it would have to examine if being fast with an answer was useful in running a foot race. It would also have to examine if there were costs that might interfere with the utility of the synthesis. The problem with these complications is, of course, that they might have their own complications. It is one thing to discover that an actual process can be categorized as a kind of process but it is another thing to be able to use the different variations of that kind of processes effectively. And as the program gained more information about different kinds of things it might have to consider a number of ‘philosophical’ considerations concerning the subject which might or might not be useful to its analysis.
Anyway, this synthesis is the basis of discovering and using an ‘objective’ which can be correlated with a goal or a subgoal. So it truly is an important method. And any example of a creative application of a concept to some other situation involves a stage of synthesis. As long as future AGI programs are written to try to evaluate how well a feature of an ‘object’ might be applied to a situation where the feature might be used effectively then this process of algorithmic synthesis would be perfectly feasible. (One of the interesting things is that it might be used in a completely artificial language where features of 'objects' could be annotated.) Jim Bromer ------------------------------------------- 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
