|Intelligent Systems And Their Societies||Walter Fritz|
The program is an evolution of the previously described ones. Now we call the "triplets" and "experiences", "response rules". We express the parts of the response rules with numbers as always, but now these numbers represent concepts. Previously, the IS used only concrete concepts; now the concepts can be concrete, composite, or abstract.
For the first time the program notes and stores rule patterns, that is, the relation between concepts of the situation side and the response side of response rules. Without this, most of the capabilities noted above are not possible (greeting by name, properties of lines, curves, and so on, and arithmetic). The selection of response rules is not just by "emotion" but by a sophisticated method.
The program first tries to select a response rule by expressing the present situation as concrete concepts, then as composite concepts and finally as abstract concepts. The IS attaches positive or negative values to all concepts of the situation of the response rules. The choice of response rules takes into account the value of each concept of the response rule that it also finds in the present situation; also of those it does not find and finally, those existing in the present situation but not in the response rule. This gives a final value for each response rule. Now the IS chooses only one rule from this list. It chooses by chance, but those rules with more "value" have more probability to be chosen.
During the sleeping period, a period of external inactivity, the system creates abstract and composite concepts and more generalized response rules. Also it forgets response rules, concepts and patterns that are not much used.
For continuous reading, like a book - continue
Jump to the top of this document /General Learner document /Artificial IS document /e-book Contents.
Copyright © New Horizons Press