| Intelligent Systems And Their Societies | Walter Fritz |
by Ganesh Mani and Leonard Uhr of the Computer Sciences Dept. of the University of Wisconsin at Madison.
CHILDLIKE's central mechanism is a perceive - reason - act - learn loop. There are two input channels, one visual and one linguistic. The system's visual input is in the form of a series of successive snapshots of the environment. This input was still simulated in 1991. The system's linguistic input is in the form of a string of words.
Both of these inputs together are what we call a situation (a snapshot of the environment). A situation has the effect of grounding words in perceptual information. Once CHILDLIKE has learned words, it continues learning using these grounded words alone. Thus it creates abstract concepts.
The goal in building this system was to create a system that can learn, with a minimum of initial knowledge, to satisfy its internal needs, learn about objects and their relations, and learn words that describe these. To do this, objects and the corresponding words are associated into concepts. The system then associates a response with changes in the visual frames (we say with changes in the situation) and changes in internal needs (similar to our response rules). The system represents these response rules in an information network, where each node is a concept and the links are the response rules. Selection of a response rule to execute is by similarity matching.
The system uses composite concepts since it extracts potentially useful information from the visual and linguistic input and aggregates them into concepts. The objective of the system is to satisfy its most pressing need, such as thirst or hunger.
For continuos reading, like a book - continue
here
Jump to the top of the CHILDLIKE document /
Artificial IS document /
e-book Contents.
Last Edited 6 Mar. 06 / Walter Fritz