We present a framework to study agent-environment systems from an information-theoretical perspective. For this, we use the formalism of Causal Bayesian Networks to model the probabilistic and causal dependencies of various system variables. This allows one to formulate a consistent informational view of how an agent extracts information from the environment, including the role of its actions as a natural part of the model. The model is motivated by increasing evidence of the importance of Shannon information for the behaviour of living organisms. We relate the model to existing views on information maximization and parsimony principles and apply it to a simple scenario demonstrating the discovery of implicit structured environment models by an agent with only a strongly limited and purely local sensorimotor embodiment. Further variations of the model are briefly introduced and discussed. The chapter concludes with an indication of relevant contributions for further research.
|Title of host publication||Information Theory and Statistical Learning|
|Editors||Frank Emmert-Streib, Matthias Dehmer|
|Publication status||Published - 2009|