A motivational system that drives the development of activity
Proceedings of the first international joint conference on Autonomous agents and multiagent systems: part 1
| Hi-index | 0.00 |
We are implementing a theory of the development of activity in intelligent agents. In our theory, hierarchies of activities develop incrementally, by learning from increasingly sophisticated interactions between the agent and its environment. Ultimately, the types of interactions an agent attempts to bring about will determine the type of knowledge that will develop. In this paper, we propose a simple motivational system that makes the decisions of what an agent attempts to do. We explore how this motivational system can manage the tradeoff between exploring the environment for the purpose of learning and exploiting the environment to attend its basic needs (the need to rest, avoid pain or damage et cetera). We demonstrate the efficacy of the motivational system in a simple hand-designed environment and show how it extends to effectiveness in a more general class of randomly generated environments implemented as Markov decision proceses (MDPs).