Apportionment of work among environment, body and brain of an agent
Carlotta Langer and Nihat Ay
Contact the author: Please use for correspondence this email.
Submission date: 25. Jan. 2021 (revised version: February 2021)
Keywords and phrases: information theory, information geometry, Planning as Inference, Morphological Computation, Integrated Information
Download full preprint: PDF (1062 kB)
An embodied agent, that performs a goal-directed action, influences its environment and gets influenced by it. Hence information flows not only among the agents control mechanisms, but also between the agent and its environment. In this article we combine different theories in order to create a framework in which we are able to closely examine the information flow among the body, brain and environment of an agent. We test this framework in a simple experimental setup. There the agents do not learn to perform a task, but we calculate the optimal behavior using the method Planning as Inference, in which the information geometric em-algorithm is used to optimize the likelihood of the goal. Then we analyze the resulting distribution using in- formation theoretic methods including measures corresponding to the concepts of Morphological Computation and Integrated Information Theory. Comparing the behavior of these measures under changing morphological circumstances highlights the asymmetric relationship between these two concepts.