My recollection is that Ben said that my summary only concerned declarative knowledge. Almost all programs effectively transform declarative statements (input) into procedural statements. The program branches in different ways dependent on the values of the input for a variable (or some effective action dependent on the input). But I consider the ability to intelligently transform a pure declarative statement into an action to be a necessary goal-seeking ability for an AGI program. The program has to wisely or correctly learn that a particular part of a statement can be correlated with an action. This would require giving the program some kind of meta awareness of its 'behavior'. Again, since all programs can transform declarative statements into effective actions through a programmed correlation between an input value for a variable and a branching of the program, I don't think that the learning model (where the correlation between some statement and some acquired or programmed behavior has to be learned) is that impossible. And again, this model can be simplified so that it can be easily tested and built from scratch. I think this is an important insight, not because it is complicated, but because it simplified an extremely subtle distinction that is totally relevant to the problem of creating contemporary AGI programs.
------------------------------------------- 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
