INTELLIGENT BEHAVIOR IN MACHINES EMERGING FROM A COLLECTION OF INTERACTIVE CONTROL-STRUCTURES

Control strategies that decompose the competence of an agent into inde pendent, task-achieving control structures are emerging as viable alte rnatives to the notion that a robot is an explicit symbol processing d evice with input and output extensions in the form of sensors and actu ators. This approach draws inspiration from ethology and the cognitive sciences, where the competence of biological creatures is seen to be a result of the successful combination of several distinct behavior pa tterns that enable them to interact effectively with their environment s. The distinction between classical approaches that view intelligent behavior as the outcome of symbolic manipulations and an alternate app roach that asserts that it is the careful integration of distributed, goal-competent control structures that lead to intelligent behavior is analyzed in detail. It is found that this dichotomy can be said to ar ise from the paradigm that is chosen to represent the knowledge at the disposal of the agent. Structurally encoded knowledge requiring an ac tive interpreter for the knowledge to be exhibited in action leads to the symbol processing paradigm for intelligent behavior, whereas the i nteractive model for knowledge representation assumes that the knowled ge at the disposal of an active agent resides within goal-competent, i nteractive control structures. A formalism is developed for representi ng the goal-directed nature of these interactive control structures wi th clear control-module boundaries. An implementation is then presente d to demonstrate some of the basic features of this methodology.