Cybernomics - An Open World of Adaptive Economic Learning

Héctor J. Lazzarini
Consejo de Investigaciones, Universidad Nacional de Rosario, Argentina

Abstract

This paper presents Cybernomics, the study and experimentation of adaptive learning of economic agents by means of the identification and significance (hierarchy-classification) of agent's behavior in spatial context, both computerized and simulated. This knowledge of the adaptive behavior patterns will allow to guide, predict and reorganize the adaptive economic thought, faced to a high complexity, dynamic and speed. Last technical antecedents of this approach are the experience in small of the test-bed systems, the deliberative planning and the classifier systems of the genetic programming. This paper has two parts, one theoretic-conceptual and the other practical-experimental.

The first part has three cybernomics principles. The first called Instrument-Sign Integration is related with the existence of the sign-image, the mediator between the stimulus and the adaptive learning answer. Under this concept, the cognitive associationists (inductive) and the organists (conceptual reorganization) theories are complemented.

The second principle, significance (hierarchy-classification) is related with the Cognitive Space (Map) of the learning and the perception (vertical-horizontal), that is needed to orientate the process of automatic selection of adaptive rules.

The third principle, the Molar Identity recognizes the behavioral models (patterns), like a whole not reducible to its elements or strings of simple code. The models are tracks of thought within the infinite combinations of the adaptive open world.

The second part presents Cybernomics I, a computerized intrumental-symbolic space under the form of grill. In the first instance of its learning, the agent tries to follow success. After the repetition of new experimental sequences in the computer arrives to a superior level of the adaptive knowledge, understand success. This object is achieved by the computer by means of an experimental mechanism of autoselection and improvement, beginning from behavioral patterns (molars) under dynamic hierarchies (classifier systems). To achieve this object the computer asserts the ``tracks" of the agent's movements for the further treatment. Most part of it is develped in memory. This ordination is based on a criterion a little different and complementary of the classificative (genetic) rules.

Spatial impacts (wall, obstacle, trash, etc.) are added to the stimulus-reinforcement and sensibility classic hulian approaches that cause that the computer learn how to select the best behaviors in reparameterized environments. The annual meetings of the AAAI Robot Competition and the displacement philosophy of movable agent of project MAIA (Artificial Intelligence Advances Model) added to the system the following ideas: the chance to make a Conceptual Cognitive Map for the agent's physical-conceptual movement and the ability to detect ``conceptual trashes" (internal detector). Those trashes have contrary functions respect to the tiles in the known Tileworld; obstruct and not help instrumentally the adaptive success. Finally cybernomic application examples in economic models and classifiers are given.