Interaction of a Neural Network Driven Agent with the Environment

			Matej Hoffmann 
                             MFF UK 


A novel perspective on cognition, as represented by the works of
H. Maturana and F. Varela, will be briefly presented. In a nutshell,
they view cognition as embodied action and emphasize the importance of
continual mutual adaptation (or structural coupling) of an agent and
environment. The talk will concentrate on the implementation of these
concepts in a simple agent driven by a neural controller, in constant
interaction with a virtual environment. Different neural architectures
and strategies of their development (evolution/learning) will be
compared with respect to the agent's performance when interacting with
the environment. The grounding problem (to what extent is the
architecture predesigned) and biological plausibility will also be
addressed.