Cognitively Inspired Anticipatory Adaptation and Associated Learning Mechanisms for Autonomous Agents

This paper describes the integration of several cognitively inspired anticipation and anticipatory learning mechanisms in an autonomous agent architecture, the Learning Intelligent Distribution Agent (LIDA) system. We provide computational mechanisms and experimental simulations for variants of payoff, state, and sensorial anticipatory mechanisms. The payoff anticipatory mechanism in LIDA is implicitly realized by the action selection dynamics of LIDA’s decision making component, and is enhanced by importance and discrimination factors. A description of a non-routine problem solving algorithm is presented as a form of state anticipatory mechanism. A technique for action driven sensational and attentional biasing similar to a preafferent signal and preparatory attention is offered as a viable sensorial anticipatory mechanism. We also present an automatization mechanism coupled with an associated deautomatization procedure, and an instructionalist based procedural learning algorithm as forms of implicit and explicit anticipatory learning mechanisms.