Intoduction to Part XI

Abstract

Although learning is not a necessary capability of an intelligent system, it is one of the particularly intriguing capabilities exhibited by many intelligent “agents”. However, much like intelligence itself, the word “learning” suggests a homogeneity of the phenomenon that is deceptive. In the great majority of cases, particularly in organisms with higher developed brains, the capability to learn is subserved by several, often interlinked mechanisms, each responsible for a particular aspect of the learning capability. Thus, to understand “learning” requires to answer questions on a variety of different levels. From the viewpoint of biology, these levels range from adaptive processes at single synapses over adaptive changes in local networks up to the highest levels where cognition, emotion and motivation controls the activation of various subsystems each of which contributes its special part to learning in the organism. From the viewpoint of information science, one would like to understand the computational principles behind the processes implemented at these different levels. And finally, from the viewpoint of engineering and robotics, one would like to use these principles to build artificial learning systems, using physical implementations that may be inspired by the biological ones, but also guided by a set of different constraints resulting from the world of silicon and artificial sensor devices.