The agents evolved by Randall Beer for active object discrimination perform an undeniably interesting task. It may seem modest from the point of view of an external human observer, but for a robot with limited sensory capability trying to work out twodimensional shapes from a ﬂux of one-dimensional data, it is indeed challenging. Already in this fact we ﬁnd a serious answer to critics who think that simple experiments in mobile robotics like this one have little relevance for understanding cognition as a natural phenomenon. Whether a task is cognitively interesting cannot be judged in a vacuum, or only by human standards, but depends on the dynamical, bodily and environmental conditions with which an agent must cope. If by his initial judgement the designer cannot foresee a trivial way to perform the task given the resources provided to the agent, then the synthesis of successful behaviour is an event from which something can certainly be learned. Such is the motivation that is intuitive for many people working within autonomous and evolutionary robotics, but which often seems to escape those who think that these disciplines are solely focused on synthesizing human level intelligence from the bottom up.
Di Paolo, E. A. and Harvey, I., (2003). Decisions and noise: the scope of evolutionary synthesis and dynamical analysis. Adaptive Behavior 11(4): 284-288.